Genotoxic Agents Produce Stressor-Specific Spectra of Spectinomycin Resistance Mutations Based on Mechanism of Action and Selection in Bacillus subtilis

ABSTRACT

Mutagenesis is integral for bacterial evolution and the development of antibiotic resistance. Environmental toxins and stressors are known to elevate the rate of mutagenesis through direct DNA toxicity, known as stress-associated mutagenesis, or via a more general stress-induced process that relies on intrinsic bacterial pathways. Here, we characterize the spectra of mutations induced by an array of different stressors using high-throughput sequencing to profile thousands of spectinomycin-resistant colonies of Bacillus subtilis. We found 69 unique mutations in the rpsE and rpsB genes, and that each stressor leads to a unique and specific spectrum of antibiotic-resistance mutations. While some mutations clearly reflected the DNA damage mechanism of the stress, others were likely the result of a more general stress-induced mechanism. To determine the relative fitness of these mutants under a range of antibiotic selection pressures, we used multistrain competitive fitness experiments and found an additional landscape of fitness and resistance. The data presented here support the idea that the environment in which the selection is applied (mutagenic stressors that are present), as well as changes in local drug concentration, can significantly alter the path to spectinomycin resistance in B. subtilis.

INTRODUCTION

While maintaining the integrity of DNA is vital to survival, the introduction of mutations is equally important for the long-term success of a population as it enables continued adaptation. In bacteria, mutations that lead to even a single nucleotide change can have an extensive impact on phenotype, such as by changing host tropism (1), altering virulence (2), and introducing antibiotic resistance (3). Point mutations can be introduced through a variety of mechanisms including the induction of mutagenic DNA repair pathways (4–7), inhibition of DNA repair (8, 9), homologous recombination (10), replication errors (7, 10–12), and direct base damage (7, 13–15). We now understand that the rate of mutations is not static and is elevated by a variety of stimuli, and the biology, genetics, and chemistry behind these mechanisms have been extensively studied (7, 16–21).

Environmental mutagens, as well as bacterial stress responses, play a key role in generating mutations that drive evolution and lead to antibiotic resistance (16, 22). Mutagens such as heavy metals (23), UV radiation (24), and antimicrobials (25–28), as well as exposure to stressful conditions such as starvation (29), have been shown to induce antibiotic resistance through mutation. Additionally, previous work has shown that different stressors can produce different types of mutations (30–34). These types of stressors have the capacity to cause mutation by directly damaging DNA through alkylation, oxidation, or cross-linking of bases, in a process termed stress-associated mutagenesis (SAM) by MacLean et al. (16). Bacterial stressors can also act through stress-induced mutagenesis (SIM) by regulating cellular genetic machinery that leads to mutations, such as the induction of error-prone polymerases or inhibition of DNA repair genes (8, 9, 16, 35). Since both of these mechanisms can act in response to a particular stressor, the spectrum of observed mutations would reflect a mixture of SAM and SIM.

Understanding the mechanisms by which bacteria develop antibiotic resistance is key to combating the current antibiotic-resistance crisis, which results in millions of resistant infections and tens of thousands of deaths annually (36). Instances of genetic mutation leading to resistance are well documented, and some of the most commonly used antibiotics have lost much of their efficacy due to the emergence of resistance-conferring mutations in a few bacterial genes. In fact, resistance to both first- and second-line drugs in M. tuberculosis is made possible by single point mutations in a number of genes, highlighting the great impact that single nucleotide substitutions can have on bacterial phenotypes (3).

Mutations that lead to resistance against ribosome-targeting antibiotics represent a major threat to our current medical practices. Spectinomycin is an aminocyclitol antibiotic which inhibits translocation of the peptidyl-tRNA from the A- to the P-site by binding to the 30S subunit (37–39). This antibiotic is especially important, as it is an effective option for treating Neisseria gonorrhoeae infections, which have become resistant to penicillin (40–42), sulfonamides (43), tetracycline (44), and, worryingly, the cephalosporin ceftriaxone (45, 46). Spectinomycin resistance via single nucleotide mutations in ribosomal protein S5 was identified previously in N. gonorrhoeae (46, 47), Escherichia coli (48, 49), and Bacillus subtilis (50), as well as other bacteria (51, 52). In B. subtilis, both the rpsB and rpsE genes that code for the S2 and S5 proteins contain multiple sites where point mutations have the potential to confer resistance. Looking at the spectrum of mutations in these two genes can be useful for understanding how environmental stressors cause spectinomycin mutations through a combination of SIM and SAM.

In both experimentally and naturally selected mutations, the final product of the selection represents only a fraction of the total mutational landscape. This landscape is defined by the initial mutagenic stress, such that different stressors will induce a different array of mutations (30–34) via combination of SIM and SAM mechanisms, depending on the context of the stress (16). This spectrum of mutations then undergoes various selective forces, including selection for antibiotic resistance (53–57) and bacterial fitness (16, 58–62), that shape the final mutational profile observed. Here, we utilize a high-throughput system employing traditional microbiological techniques in conjunction with next-generation sequencing (NGS) to demonstrate the impact of an array of genotoxic stressors on the spectrum of spectinomycin-resistance mutations, characterizing the mutational spectrum associated with each genotoxic agent within the rpsE gene of B. subtilis. We find that each stressor generates a unique set of mutations, some of which are reflective of their respective mechanisms of DNA toxicity and suggest SAM, whereas others likely arise from SIM mechanisms. As part of this effort, we also identify novel spectinomycin-resistance mutations in both the rpsE and rpsB genes of B. subtilis. From the full spectrum of stressor-specific mutations, selective processes narrow down the mutations to viable mutants. Utilizing a bacterial competition assay combined with a sequencing approach, we find that, consistent with previous work, the fitness of various mutants is highly dependent on antibiotic concentration. Our results demonstrate that the full spectrum of available spectinomycin-resistance mutations is narrowed down to a few mutations during multiple selection events. Thus, the final mutants that emerge are shaped by both the initial mutagen/stress and the subsequent level of antibiotic selection pressure.

RESULTS

Genotoxic agents induce a spectrum of unique and treatment-specific resistance mutations.

Past work indicates that exposure to mutagens and other forms of stress can cause genotoxicity and induce mutations that lead to the development of resistance. We tested the propensity of an array of genotoxic conditions, each inducing DNA damage through a different mechanism, to generate spectinomycin resistance in the wild-type (WT) background of B. subtilis strain 168 compared to no-treatment controls. As growth phase is an important factor in stress response and mutagenesis (63–65), we exposed cells to mutagenic agents for 30 min during logarithmic phase or for 3 h during stationary-phase growth before removing the stressing agent and plating on spectinomycin selection medium (100 μg/ml) to determine the number of resistant colonies generated. The benefit of this experimental design is that the stressor is experimentally and temporally separate from the following spectinomycin selection. The short exposure of stress prevents cells from undergoing multiple rounds of division and selection. Treatment concentrations were determined by dose titration experiments and selected based on mutagen exposure levels that were able to potentiate mutagenesis while minimizing bacterial killing. The mutational frequency of each stressor is likely the result of an agent-specific mechanism of action that can also change in a concentration-dependent manner for each of the agents (21). For example, antibiotics like ciprofloxacin may induce DNA mutations through either the direct mechanism on DNA gyrase/topoisomerase or via reactive oxygen species (ROS)-induced damage. Here, stressor concentrations were determined using a dose titration and selecting concentrations that induced the greatest frequency of mutants. These experiments provided both the extent of mutagenicity of the different treatments and also generated a compendium of thousands of individual resistant colonies that we used to elucidate the spectrum of mutations associated with each treatment.

The frequency at which spectinomycin-resistant colonies were generated was dependent on both the type of mutagenic agent administered as well as the growth phase (Fig. 1A). Mitomycin C, ciprofloxacin, and UV radiation all increased the frequency at which resistant colonies developed compared to untreated cells across growth phases, while hydrogen peroxide and cobalt chloride caused an increase in resistant colony frequency only during logarithmic growth (Mann-Whitney U test, P < 0.05) (Fig. 1A). In addition, untreated cells grown to stationary phase exhibited an increase in the formation of resistant colonies compared to untreated cells grown within logarithmic phase (Mann-Whitney U test, P < 0.05), likely reflecting the stress incurred by extended growth and the known phenomenon of stationary-phase mutagenesis in B. subtilis (63). These data suggest that stressors associated with different mechanisms of DNA damage are capable of inducing spectinomycin resistance in B. subtilis to different degrees depending on both the type and timing of exposure. The fact that stressors further elevate mutations during stationary phase compared to untreated cells likely indicates that mutations derived from SAM mechanisms are additive to SIMs related to stationary phase.

FIG 1.

Genotoxic agents induce a spectrum of unique, treatment-specific resistance mutations. (A) Mutation frequency of spectinomycin resistance in WT B. subtilis treated with various stressors during logarithmic- or stationary-phase growth. Mutation frequency was calculated by dividing the number of spectinomycin-resistant colonies by the total number of cells. Significant differences between untreated and treated cells were determined using the Mann-Whitney U test (***, P ≤ 0.0001). Horizontal LOD line represents the limit of detection. (B) Spectrum of nucleotide mutations in the rpsE gene of the spectinomycin-resistant mutant colonies of WT B. subtilis. Notably, there were no mutations identified in the no-treatment condition during logarithmic phase, as this condition produced very few or no resistant colonies. In lieu of the resistant colonies, logarithmic-phase cells were plated on nonselective medium and 500 of these nonresistant isolates were sequenced. (C) The relative abundance of nucleotide mutations in the rpsB gene from 90 spectinomycin-resistant mutant colonies compared to the relative abundance of mutations from the rpsE gene from 480 colonies (including the 90 colonies displaying the rpsB mutations on the left), all of which were from UV-treated cells. For B and C, each treatment represents ∼500 colonies collected from 3 to 12 biological replicates.

We utilized Sanger and whole-genome sequencing on a subset of colonies to determine the genetic basis of resistance in these mutagenized bacteria. These strategies identified base substitutions in two genes, rpsE and rpsB, encoding the ribosomal proteins S5 and S2, respectively (Table S1 in the supplemental material), aligning with a previous report that showed spectinomycin resistance in B. subtilis can result from a single amino acid replacement in the ribosomal protein S5 (50). More unique mutations were found in the rpsE gene compared to rpsB (Fig. S1 and S2), therefore we chose to examine the spectrum of mutations in rpsE across all previously generated spectinomycin-resistant colonies using an amplicon-based, next-generation sequencing strategy.

We used the rpsE gene as a model to understand how different stressors can induce spectinomycin resistance through different routes of mutation, leading to unique, stress-specific spectra of mutations. Previous work in both prokaryotic and eukaryotic models has suggested that some mutagens have mutational signatures (66). We aimed to uncover the mutational spectrum in rpsE by isolating and sequencing thousands of individual spectinomycin-resistant colonies derived from cultures exposed to different treatments. In order to identify the largest possible range of mutants, we employed a low selective pressure of less than 2× MIC (spectinomycin) and grew colonies on uncrowded plates, anticipating that this would enable recovery of isolates with both common and rare resistance mutations while reducing potential negative fitness impacts of individual mutations. To identify the mutational spectrum of various stressors, we performed rpsE amplicon sequencing on an Illumina MiSeq to an estimated average sequencing depth of roughly 1,500 reads per colony on a library of ∼5,000 individual colonies from 10 different treatment conditions (∼500 colonies per condition), representing at least 6 replicates per condition.

Sequencing revealed that each treatment induced a unique set of nonsynonymous mutations, most commonly resulting in single amino acid changes (Fig. 1B). In all of the conditions, the relative abundance of colonies with an identified rpsE mutation was less than 100%, indicating that mutations occur outside of this gene, and the frequency at which a colony was found to contain an rpsE mutation varied between conditions. For example, it appears that the rpsE gene is more responsible for spectinomycin resistance in mitomycin C-treated cells (71.72%) compared to those exposed to UV radiation (12.4%) (Fig. 1B). It is likely that the low percentage of rpsE mutations in some conditions resulted from a higher frequency of mutations in other genes, such as rpsB. This finding highlights the fact that, given multiple gene targets, different stressors will repeatedly and preferentially target certain genetic loci.

Mutation spectra correspond to stressor-specific mechanisms of mutagenesis.

Examination of the spectrum of mutations corresponding to different stressors reveals both expected patterns of mutation based on known mechanisms of mutagenesis and novel, broad patterns of mutation (Fig. 1B). The DNA damage mechanism of mitomycin C (15), UV (14, 67), and H₂O₂ (68, 69) has been previously described and, thus, we were able to predict what mutational spectrum these stressors would produce.

Mitomycin C, an antitumor drug, induces DNA damage by forming interstrand cross-links of guanosine residues at CpG sites (15, 70, 71). The data here reveal that the majority of mitomycin C-induced spectinomycin-resistance mutations occur at a CG (nucleotides 88 and 89), where it accounted for over 50% of the rpsE mutations during both logarithmic- (68.6%) and stationary-phase (61.41%) growth. However, in all other treatments, the G89C mutation was only present in a maximum of 8.54% of rpsE mutants. The mitomycin C mutation spectrum displays how patterns of mutation caused by DNA damage can be stress-specific and, in the case of mitomycin C, is reflective of the particular DNA damage mechanism of the drug. Thus, in the case of mitomycin C, SAM likely plays a greater role than SIM.

UV radiation also has a unique DNA-damaging mechanism that commonly targets adjacent pyrimidine bases leading to the C-to-T base substitutions known as UV signature mutations (14, 67, 72). A mutation in the rpsE gene was relatively rare among colonies generated by UV treatment (12.43% in logarithmic phase; 18.22% in stationary phase). This could reflect a lack of adjacent pyrimidine bases that could result in a productive mutation in this gene. To determine whether this motif was responsible for the majority of the resistance observed in UV-treated cells, we Sanger sequenced the rpsB gene from 90 UV-treated colonies. Roughly 68% (62 colonies) of the colonies contained a C-to-T mutation characteristic of UV damage (Fig. 1C). These findings illustrate that the rpsB gene is a more common target for spectinomycin-resistance mutations in the context of UV radiation exposure, likely due to the susceptible nucleotide sequence (adjacent cytosine bases) that lends itself to the UV signature C-to-T mutation.

These data suggest that bacterial evolution and development of antibiotic resistance through SAM is shaped by a combination of the mutagenic stress, the mechanism of DNA damage, the nucleotide composition of the resistance loci, and the potential for the DNA damage to induce the type of mutation that will lead to resistance. Within the genome of B. subtilis, and even within rpsE and rpsB, there are numerous sites of adjacent pyrimidines. However, out of all these sites that are subject to UV damage, only the C74 site in rpsB is strongly selected for, likely based on its ability to provide resistance in a way that does not compromise cell viability. It is likely that some rpsE mutations were not detected due to epistatic or lethal mutations elsewhere in the genome, especially under highly mutagenic conditions. Ultimately, UV is another example where SAM is likely the predominant mechanism of mutagenesis leading to observed mutations.

We also found an observable mutation signature in the case of hydrogen peroxide, analogous to observations made regarding mutations of the supF gene of E. coli (69, 73). Akasaka et al. found that the mutational spectra of hydrogen peroxide were predominantly G:C-to-C:G (40%) and G:C-to-T:A (37.14%) transversions (69, 73). We identified a similar profile of mutations in the B. subtilis rpsE gene (Table S2), finding that a large portion of the mutations were G-to-C (23.9%) and G-to-T (23.4%) transversions; however, we also observed a substantial number of G-to-A transitions (32.2%) (Fig. 1B). Interestingly, when the supF gene was exposed to hydrogen peroxide and then passaged through simian (CV-1) cells, G:C-to-A:T mutations represented 43.5% of the overall mutations (68). The high abundance of mutations at guanine residues in both our data and previous studies is likely reflective of the mechanism of hydrogen peroxide damage, which forms hydroxyl radicals known to cause 8-oxo-dG lesions (74).

We also examined the mutational spectrum of stressors for which mutational spectra were unknown and could not be inferred by a distinct mechanism of DNA damage. Sequencing mutants generated by exposure to stationary-phase stress, CoCl₂, and ciprofloxacin provided an opportunity to observe novel mutational spectra and gain insights into new mechanisms of mutagenesis.

Previous research into B. subtilis stationary phase mutagenesis revealed that subpopulations of stationary-phase cells undergo adaptive mutagenesis (63), but there has not been significant investigation into the spectrum of mutations associated with this condition in B. subtilis. Overall, we found that stationary phase was not associated with an overrepresentation of a specific mutation, but rather there was a mostly even distribution among 4 different mutations. Such a distributed pattern was also observed in other treatments, such as ciprofloxacin during stationary phase. In the case of mitomycin C treatment, in both growth phases the overall mutation spectrum is not an even distribution of multiple mutations but is dominated by a mutation linked directly to a specific form of DNA damage induced by the drug (Fig. 1B).

Cobalt (Co) in the form of cobalt chloride (CoCl₂) has been shown to be mutagenic both in bacterial and mammalian cells (75, 76). The generation of DNA-damaging reactive oxygen species (ROS) has been proposed as a method by which metal ions induce toxicity and potentially mutagenesis (77, 78), but there is not substantial evidence to link Co-induced mutagenesis to ROS in bacteria (78). However, we observe that the most common mutation in the rpsE gene of B. subtilis treated with CoCl₂ is G-to-T (46.2%) (Fig. 1B), which could potentially be a signature of 8-oxo-dG mutations caused by ROS (79). Interestingly, this is an even higher rate of G-to-T mutations than the hydrogen peroxide treatment, which would be expected to exhibit more signatures of ROS-associated DNA damage. A potential alternate mechanism is that Co may cause toxicity through inhibition of DNA repair, independent of oxidative stress, as has been suggested to occur in E. coli (80). However, based on the data presented here, we are unable to link Co mutagenesis to a distinct mechanism, though we do show that CoCl₂ has the potential to induce base substitution mutations that result in antibiotic resistance through a unique spectrum of mutations.

Antibiotics are known to cause mutagenesis through activation of both SOS-dependent (28) and independent pathways (81), which may be related to ROS production (25, 28). Treatment of B. subtilis with ciprofloxacin led to an increase in the formation of spectinomycin-resistant colonies (Fig. 1A). The predominant rpsE mutation of ciprofloxacin treatment is A67G in both log phase and stationary phase, with this mutant making up 21.94% and 17.41% of the relative abundance of resistant mutants, respectively (Fig. 1B). However, while 80.61% of the rpsE mutant reads are A67G in log-phase cells treated with ciprofloxacin, A67G was only 29.16% of the rpsE mutations in ciprofloxacin-treated stationary-phase cells (Fig. 1B). This suggests that ciprofloxacin treatment has a variable mutation spectrum that depends on growth phase, and it is likely that the combined stress of stationary-phase growth in addition to ciprofloxacin-induced stress is responsible for a more diverse mutational spectrum during stationary-phase growth. This highlights the importance of evaluating the mutational spectrum of mutagens at different growth phases and understanding the combinatorial impacts of stressors. It is also possible that the divergent signature observed between stationary- and logarithmic-phase cells treated with ciprofloxacin results from different rates of replication fork stalling-induced double-strand breaks that may occur between dividing and nondividing cells, potentially altering the relative contributions of SAM and SIM mechanisms.

Bacterial evolution studies have revealed that, like many other organisms, bacteria exhibit transition bias, with transition mutations overrepresented relative to the unbiased T_i:T_v (82, 83). Of the spectinomycin-resistance mutations in the rpsE gene, the ratio of transitions to transversions is dependent on the type of mutagenic stress administered to the bacteria (Fig. S4A). While ciprofloxacin and hydrogen peroxide treatments, as well as stationary-phase stress, induced a higher ratio of transitions to transversions in rpsE than expected in an unbiased scenario (T_i:T_v > 0.5), mitomycin C, UV radiation, and cobalt chloride treatment instead all resulted in further bias toward transversions (T_i:T_v < 0.5) (Fig. S4A). Interestingly, across all the possible observed mutations, there were nearly 4 times more detected (37) transversion mutations compared to transitions (9). It has been proposed that at the amino acid level, transversion mutations are more likely to cause a change in the biochemical properties of amino acids and that transition mutations are more conservative (84, 85). Therefore, it may be logical to assume that a resistance-conferring mutation would be a transversion. However, a more recent meta-analysis found transition mutations to be only slightly more conservative than transversions (86), and thus there may be alternative explanations for the increased abundance of transversions observed here. Due to the different mutagenic pathways in each of the treatments, not all mutational spectra reflect a transversion bias, highlighting the unique stressor-specific pathways that lead to antibiotic-resistance mutations.

Location of mutations in ribosomal proteins S5 and S2.

Alterations of the loop 2 region of the S5 protein have been shown to impact spectinomycin sensitivity in E. coli by disrupting the binding of the drug to the ribosome (87, 88). Sanger sequencing of the entire rpsE gene revealed that the observed mutations were strictly limited to a small region of the protein and, in agreement with previous studies, these mutations were located along the loop 2 region of the S5 protein (Fig. S5A). We identified 63 unique rpsE mutant alleles with single, double, or triple nucleotide mutations (Fig. S3). Of these 63 mutations, only 2 have been previously described in B. subtilis (50, 89). We observed 6 different mutation sites in the S2 protein, but the most commonly mutated sites were at amino acid positions 22 and 25 (Fig. S5B). The close proximity of these two sites and their high frequency of mutation in resistant colonies suggest that this region of the S2 protein may have an impact on the binding of spectinomycin to the ribosome. These localization patterns suggest that selection for mutations is dependent not only on the possible mutations generated by different stressors, but also that these mutations occur in sites that will provide a physiologic advantage such as resistance.

Impact of DNA repair genes on the frequency and spectrum of mitomycin C-induced mutagenesis.

Based on the data presented above, the sequenced mutants likely result from variable combinations of SIM and SAM mechanisms. In order to ascertain the contributions of each mechanism, we decided to observe how disruption of processes known to be associated with SIM, including DNA damage repair processes and the SOS response, impacted the spectrum of spectinomycin-resistant mutants generated through mitomycin C treatment. Specifically, we tested the mutation frequency and mutational spectrum of a library of previously validated mutants targeting processes related to DNA maintenance and repair (90). Several mutants lacked genes involved in the SOS response, including yhaO, yobH, yozK, uvrX, uvrA, uvrB, uvrC, recA, and dinB, which are all regulated by LexA, the transcriptional repressor of the SOS regulon (91, 92). Other mutants contained knockouts of genes involved in DNA maintenance and repair independent of the SOS response, including radA, adaA, sbcD, recN, radC, mutSB, exoAA, and mfd.

In WT B. subtilis, we had observed that the spectrum of rpsE spectinomycin-resistance mutations arising from mitomycin C exposure was dominated by a particular mutation of a CpG site, corresponding to the interstrand cross-linking that mitomycin C induces at these sites (Fig. 1A and B). Previous work has suggested that such interstrand cross-links are repaired via the nucleotide excision repair pathway (NER) or by homologous recombination (70). Mitomycin C is also an alkylating agent capable of generating mono-adducts at the N² and N⁷ positions of guanine, which in B. subtilis are also repaired through NER, including both the UvrABC and MrfAB pathways (70).

Utilizing the mitomycin C mutagenesis conditions used previously during logarithmic growth, we found that the deletion of genes involved in DNA maintenance and repair had vastly different effects on the frequency of formation of spectinomycin-resistant colonies (Fig. 2A). DNA repair is a vital part of counteracting the mutagenic impacts of various genotoxic stressors. Compared to the frequency observed in WT, the ΔadaA, ΔyhaO, ΔrecN, ΔmutSB, ΔuvrA, ΔuvrB, and ΔuvrC deletions all significantly increased the frequency at which resistant colonies formed as a result of mitomycin C treatment (Mann-Whitney U test, P < 0.0001). Surprisingly, most of the ΔrecN, ΔuvrA, ΔuvrB, and ΔuvrC colonies that formed on the spectinomycin selection plates could not be regrown in liquid medium under selection, suggesting that these colonies were inviable or did not contain a true resistance mutation. In contrast, resistant colonies from the ΔadaA, ΔyhaO, and ΔmutSB deletion strains had high rates of regrowth under selection compared to the colonies derived from ΔrecN, ΔuvrA, ΔuvrB, and ΔuvrC strains, suggesting a heritable, genetically encoded resistance phenotype. Equally important are those genes that were shown to reduce the mutagenic impacts of mitomycin C treatment; strains with deletions in radA (a recA paralog), sbcD, yobH, yozK, uvrX, exoAA, mfd, and dinB all had reduced mutation frequencies compared to WT B. subtilis when treated with mitomycin C (Fig. 2A). In fact, many of these deletion strains did not produce any spectinomycin-resistant colonies after exposure to mitomycin C, nor did they not show increased sensitivity to mitomycin C. Our findings are in agreement with previous studies of mfd and dinB that have shown these genes are necessary for mutagenesis and the development of antibiotic resistance (4, 8). The knockout strain lacking radC, which encodes a protein repair homologue (93), had an increased baseline mutation frequency but did not display an increase in resistant mutants when exposed to mitomycin C. The decreased mutagenicity of strains lacking radA, sbcD, yobH, yozK, uvrX, exoAA, mfd, and dinB implies that these genes may play a role in the process of evolution and the development of antibiotic resistance through SIM.

FIG 2.

Impact of DNA repair genes on the frequency and spectrum of mitomycin C-induced mutants. (A) Mutation frequency of spectinomycin resistance in DNA repair-deficient B. subtilis treated with 100 ng/ml mitomycin C. Mutation frequency was calculated by dividing the number of spectinomycin-resistant colonies by the total number of cells. Bars represent mean ± standard error of the mean (SEM) (n = 3 minimum). Significance between mitomycin C-treated and untreated WT, and between mitomycin C-treated WT and mitomycin C-treated DNA knockouts was determined using the Mann-Whitney U Test (***, P ≤ 0.0001). Horizontal LOD line represents the limit of detection. (B) Spectrum of nucleotide mutations in the rpsE gene of the mutant colonies from WT and DNA repair-deficient strains of B. subtilis treated with mitomycin C. Each treatment represents ∼500 colonies collected from 3 to 12 biological replicates.

The three genes whose knockouts increased mitomycin C-induced mutation rates and could be regrown under selection, adaA, yhaO, and mutSB, are responsible for DNA repair in ways that could reduce DNA damage by mitomycin C. The adaA gene codes for a methylphosphotriester DNA methyltransferase and is part of the adaptive response to DNA alkylation in B. subtilis. The AdaA protein is the transcriptional regulator of alkA, which encodes a 3-methyl glucosylase that can remove 7-meG lesions, such as those formed by mitomycin C (91, 94, 95). Therefore, deletion of adaA may lead to a higher frequency of mutant formation by shifting cells to a more mutagenic form of DNA repair. The yhaO gene, which is homologous to the DNA endonuclease gene sbcD, has been shown to interact with the nuclease SbcC and may play a role in the repair of interstrand cross-links (91, 96–98). Finally, mutSB encodes the endonuclease MutS2, which may play a role in homologous recombination and potentially impact the mutational spectra in B. subtilis (99) through the involvement of this process in the repair of mitomycin C-induced damage (99, 100). That increases in rates of mutant colonies appearing in B. subtilis strains deficient in genes thought to be directly involved in repair of mitomycin C-associated alkylation damage (adaA, yhaO, and mutSB) suggests that these genes are involved in limiting SAM rather than introducing mutations through SIM, which perhaps explains why the baseline mutation frequency was not increased when compared to the WT in any of the three mutants (Mann-Whitney U test, P > 0.05).

To determine the impact of the ΔadaA, ΔyhaO, and ΔmutSB deletions on the spectrum of mutations, we sequenced the rpsE gene of the spectinomycin-resistant colonies formed by mitomycin C mutagenesis in these mutants. The frequency at which a colony was found to contain an rpsE mutation was decreased in each of the deletion strains compared to WT, suggesting that mutations are more likely to form in other regions of the genome, such as in rpsB. Consistent with observations of the WT strain, the dominant rpsE mutation observed in these deletion strains was the G89C mutation, which results in an arginine to proline amino acid substitution at position 30 (Fig. 2B), suggesting that direct base damage is still the chief mechanism of mutagenesis. While this dominant mutation remained the same, the complement of other mutations appeared to differ from the WT strain. Compared to the WT, the ΔadaA- and ΔyhaO-derived colonies had a higher proportion of mutated reads containing double mutants with both a transition and transversion mutation in the same copy of rpsE: 0.67% for WT versus 11.92% and 7.26% for ΔadaA and ΔyhaO, respectively (Fig. S4B). These data suggest that lacking adaA, yhaO, and mutSB increases mutagenicity during mitomycin C exposure, and while the mutation spectrum was still dominated by a particular G-to-C mutation, the array of other mutations appeared to be changed by the lack of these DNA repair genes. Thus, these DNA repair systems play an important role in mutagenesis and the overall spectrum of mutations.

Growth phenotypes of selected spectinomycin-resistance mutations.

Determining the functional impacts of the spectrum of mutations is critical for understanding their impacts on fitness and ultimately how they are selected for. Before a mutation results in a viable and antibiotic-resistant cell, the bacterium must be able to tolerate the mutation and overcome physiological barriers to survival and proliferation. In addition, relative fitness of the mutation is critical to the establishment of the newly developed mutant in microbial environments with and without antibiotic selection. We aimed to quantify the potential fitness effects of different mutations to find out whether different mutations might be more suitable for the development of resistance and which mutations would introduce physiological barriers to success. To better understand the physiologic significance of the spectinomycin-resistance mutations observed in our data, we isolated 11 unique mutants containing either single or double mutations in the rpsE gene (Table 1), which included 7 of the top 10 observed rpsE mutations in Fig. 1B.

TABLE 1.

Growth dynamics of competition strains^a

Strainb		MIC90 (μg/ml)	Cold sensitivity	Doubling timeb			Time to exit lag phase
AA	BP			Time (min)	95% CI	Significance	Time (min)	95% CI	Significance
WT	WT	62.5	No	27.53	22.28– 34.93		136.58	135.4–137.8
R30P	G89C	2,000	No	36.18	22.88–65.57	ns	254.72	240.0–269.5	****
K26N	A78C	>2,000	No	33.02	27.65–40.00	ns	158.86	154.0–163.7	*
K26N	A78T	500	No	26.55	20.67–34.56	ns	171.15	165.7–176.6	***
G28C	G82T	500	No	35.05	26.30–48.83	ns	194.03	185.4–202.6	****
A22P	G64C	>2,000	No	39.69	32.34–83.55	***	211.81	185.9–237.7	****
G28S	G82A	250	No	28.06	25.31–31.20	ns	133.37	131.3–135.4	ns
G28D	G83A	500	No	44.48	35.07–58.87	***	148.64	138.7–158.6	ns
K23E	A67G	2,000	Yes	40.37	32.34–51.49	*	268.2	258.4–278.0	****
G28V	G83T	>2,000	No	33.87	31.36–36.69	ns	223.74	221.3–226.2	****
G28D, G51C	G83A, G151T	>2,000	No	51.16	41.28–65.95	****	239.16	236.1–242.3	****
V24I, G28S	G70A, G82A	>2,000	No	34.40	28.99–41.40	ns	190.62	186.7–194.6	****

^aGrowth curves were used to calculate both doubling time and the length of time for stationary-phase cells to exit lag phase and enter into logarithmic growth. Significant differences in doubling time and time to exit lag phase between each strain and WT were determined using ANOVA; *, P < 0.05; ***, P < 0.001; ****, P < 0.0001; ns, not significant. The MIC is the MIC₉₀ of spectinomycin and cold sensitivity was determined by measuring growth at 20^oC.

^bAA, indicates the amino acid mutation(s); BP, indicates the nucleotide mutation(s); CI, confidence interval.

We then determined that the mutations had various impacts on spectinomycin resistance, growth rate, and delay in onset of logarithmic growth phase (length of time it took a culture to reach optical density at 600 nm [OD₆₀₀] = 0.2) (Table 1). The MIC₉₀ of spectinomycin for WT B. subtilis was 62.5 μg/ml, while the various mutant strains had MIC₉₀ values ranging from 250 μg/ml to >2,000 μg/ml (Table 1). The doubling time of WT B. subtilis was determined to be 27.53 min, while the doubling times of the mutant strains ranged from 28.06 up to 51.16 min (Table 1). In general, growth rates were fairly similar across mutants, but there were significant differences in the time it took strains to exit lag phase and enter logarithmic growth. We found that most strains had a significant delay in reaching OD₆₀₀ = 0.2 compared to WT (ANOVA, P < 0.05). These findings show that within the spectrum of possible resistance mutations, some may be better suited for growth under nonlimiting conditions by having higher growth rates or the ability to reach exponential growth in less time, while others would be well-suited to survive higher antibiotic exposures due to increased MIC₉₀ values. Previous reports of ribosomal protein S5 spectinomycin-resistance mutations in E. coli noted that cold sensitivity was associated with these mutations (87, 88, 101, 102). Indeed, we found that one of the mutations, G27D, was unable to grow at 20°C, indicating cold sensitivity (Table 1). Taken together, these physiological differences in growth and resistance may play a role in shaping the spectrum of mutations observed under different conditions or environments.

Determining the fitness of mutations through competition assays under a gradient of spectinomycin concentrations.

When we initially conducted the mutagenesis assays, we noticed that the colonies that formed on spectinomycin selection plates were variable in size, suggesting that mutations may impact the relative fitness of the bacteria. To model the potential selective pressures or benefits provided by altered growth kinetics, we utilized bacterial competition assays in the presence and absence of spectinomycin. Twelve strains of B. subtilis (the 11 unique spectinomycin-resistant mutants that were previously isolated and the WT background strain) were chosen for this experiment. Competition experiments were performed at spectinomycin concentrations that matched the MIC of the WT background (62.5 μg/ml, or “low”), the lowest MIC of the mutant backgrounds (250 μg/ml, or “medium”), and below the highest MIC of the mutant strains (1,000 μg/ml, or “high”). To initiate the experiment, multiple replicates of each strain were grown overnight and then mixed at approximately equal proportions. During initial growth, the microbial mixture was sampled at 0, 3, 6, 12, and 24 h. Cultures were passaged every 24 h, and the mixture was again sampled at the 48- and 72-h time points. Using the Nanopore MinION sequencing technology, we were able to rapidly identify the relative abundance of each strain at the various time points to determine how the level of antibiotic selection impacted the fitness of various mutant strains during competition.

While all the mutations tested were shown to confer some level of spectinomycin resistance, their ability to outcompete other strains was dependent on the level of selection they encountered. Over a short time period of 48 h, with no antibiotic selection, the relative abundance of mutations did not change drastically, although the abundance of the strains containing the G83A, A67G, and G83A-G151T double mutation were significantly decreased at the 48-h time point compared to the 3-h time point (stepdown Bonferroni, P < 0.0001) (Fig. 3A). All three of these decreased-abundance mutants were shown to have a decreased growth rate compared to WT in the absence of drug (Table 1). Additionally, the A67G and G83A-G151T mutants were shown to have significantly lengthened exits from lag phase, possibly explaining their decrease in abundance (ANOVA, P < 0.01) (Table 1).

FIG 3.

Competition of base substitution mutants over a gradient of selection. (A) Stacked bar plots of the relative abundance of each strain within a mixed community at times 0, 3, 12, 24, 48, and 72 h under spectinomycin selection concentrations of 0, 62.5, 250, and 1,000 μg/ml. Cultures were passaged to an OD₆₀₀ of 0.01 immediately after sampling of the 24- and 48-h time points. Plots represent the average relative abundance ± SEM (n = 4). Determination of statistical differences is described in the Materials and Methods section. (B) Growth rates of the 12 strains used in the competition experiments performed at each of the concentrations shown above in panel A (0, 62.5, 250, and 1,000 μg/ml spectinomycin). Points represent mean ± SEM (n = 6 minimum). Cell growth over time was determined by measuring OD₆₀₀ at 30-min time intervals. (A and B) The relative abundance plots in (A) are separated by drug concentration, and the line plots (B) correspond to the drug concentrations in the relative abundance plots shown directly above each plot.

At the 62.5 μg/ml spectinomycin concentration, reflective of the MIC for the WT B. subtilis strain, we found there were substantial shifts in the relative abundance of the mutants (Fig. 3A). At 48 h, the first time point after the first passage, there was a significant expansion of the strain containing the G70A-G82A double mutation (stepdown Bonferroni, P < 0.0001) and a significant reduction in all other mutants (stepdown Bonferroni, P < 0.0001), except those containing the G64C or A78C mutations, which did not significantly change in relative abundance (stepdown Bonferroni, P > 0.05) (Fig. 3A). This trend was again observed at the 72-h time point, where the G70A-G82A strain had expanded further to dominate the population, making up roughly 81% of the sequence reads (standard deviation [SD] ±1.77%). While there was a separate mutant strain containing the G82A mutation included in the competition, there was none containing G70A alone. The double mutant outcompeted the G82A mutant strain, indicating that this double mutation is likely beneficial under these conditions.

While we saw the sole dominance of the G70A-G82A double mutant strain at the low antibiotic concentration, this trend did not hold true for the medium concentration of 250 μg/ml. Under these conditions, the relative abundance of most strains began to diminish at 3 h and continued until the point of extinction by 72 h in all strains except those containing the A78C, G64C, G70A-G82A, and A67G mutations (stepdown Bonferroni, P < 0.0001) (Fig. 3A). By 48 h, the community had undergone a significant shift to mainly the A78C strain and the G70A-G82A double mutant. This trend was sustained at 72 h, although the relative abundance of the A67G and G64 strains still persisted at near baseline levels.

At the highest antibiotic concentration of 1,000 μg/ml, we observed a further change in the outcome of the competition experiment. Several of the strains, including those containing the G83A, G82A, G89C, A78T, G82T, and G83T mutations, as well as the wild-type strain, experienced a significant reduction in abundance within the first 24 h of growth at this high concentration of spectinomycin (stepdown Bonferroni, P < 0.0001). Of these strains, five of the seven had MIC₉₀ values that were below the 1,000 μg/ml selective concentration, suggesting they were unable to compete due to an insufficient level of resistance (Fig. 3A). However, two of the lost strains, G89C and G83T, had MIC₉₀ values of at least 2,000 μg/ml, suggesting that a lack of ability to grow under the high drug concentration was not the key factor, but rather they were unable to outgrow the other strains. This possibly suggests a growth defect under antibiotic stress compared to the dominant mutants. On the other hand, during the first 24 h of growth, the A78C and G64C strains expanded to become a majority of the population. The first passaging after the 24-h time point proved to have a significant effect on the population, allowing for further expansion of the A78C strain that neared significance (stepdown Bonferroni, P = 0.064) followed by a reduction in the G64C strain (stepdown Bonferroni, P < 0.0001). The final 72-h time point marked both continued prevalence of the A78C strain as well as a significant expansion of the A67G strain (stepdown Bonferroni, P = 0.0061), which had maintained a foothold in the total population throughout the previous time points. This final time point also marked a final reduction of the G64C mutant and the G70A-G82A double mutant (stepdown Bonferroni, P < 0.0001), resulting in a population that consisted nearly entirely of the A78C and A67G strains. The fitness of the A78C mutation at this high concentration mirrors its activity in the medium drug concentration, suggesting fitness of this mutation is enhanced at increased levels of spectinomycin selection. Spectinomycin dependence derived from a mutation in the rpsL gene was previously reported by Henkin et al. (103), however, whole-genome sequencing of the strains used in our competition assay did not reveal any mutations in the rpsL gene or other regions of the genome.

In order to understand the fitness of different mutants across spectinomycin concentrations, we performed growth assays at each antibiotic concentration used in the competition experiment (Fig. 3B). Antibiotic concentration had a dramatic effect on doubling time in the mutant and WT strains. However, within an antibiotic concentration, the doubling of time of mutants was fairly similar, with the exception of G83A-G151T at no treatment, G82A at 250 μg/ml, and G82T, G83T, and G83A at 1,000 μg/ml, which all had significantly decreased doubling times compared to other mutants (ANOVA, P < 0.05) (Fig. 3B, Table S3). Since the doubling time of most mutants was not able to explain the differences in fitness during the competition assays, we next looked at other growth parameters. We found that the most successful mutants at each antibiotic concentration were able to enter logarithmic phase growth more quickly than those they outcompeted. When we compared the time it took strains to exit lag phase, determined by the time it took strains to reach an OD₆₀₀ = 0.2, we found that at the 0 μg/ml drug concentration, those strains with the greatest latency in exit from lag phase times performed the poorest in the competition experiment (Fig. 3A and B). Strains G83A-G151T, G83T, A67G, and G70A-G82A all had significantly extended times to exit lag phase compared to their more successful counterparts (ANOVA, P < 0.05) (Fig. 3B, Table S3). The addition of 62.5 μg/ml spectinomycin challenge during growth increased the time it took many of the mutants to exit lag phase, but the most successful mutant, G70A-G82A, did not experience any increase in the time it took to exit lag phase (Fig. 3B, Table S3). At the medium drug concentration of 250 μg/ml spectinomycin, we observed that the highly successful A78C and G70A-G82A strains had the shortest times to exit from lag phase (Fig. 3B, Table S3). At the highest spectinomycin concentration of 1,000 μg/ml, the A78C mutant was one of the most successful and again had the shortest time to exit lag phase (Fig. 3B, Table S3). Interestingly, the other mutant that was able to successfully compete at 1,000 μg/ml, A67G, did not have a significant advantage in time to exit to lag phase. However, A67G did have one of the fastest doubling times at 1,000 μg/ml, suggesting that the time to exit from lag phase was not the only determining factor in fitness at higher concentrations of spectinomycin. These growth assays suggest that a combination of factors, including MIC₉₀, doubling time, and the length of time to exit lag phase contributes to the fitness of mutations at different drug concentrations.

The initial strategy employed to observe the mutational spectrum of different mutagens provided insight into the frequency of different mutations at a relatively low level of selection (100 μg/ml spectinomycin) and little to no competition between isolates. While this strategy allowed us to gain a more complete picture of the mutational landscapes (Fig. 1B), it does not reflect the process of selection that occurs in natural bacterial populations that are exposed to stressors in which mutants must compete. Using nanopore sequencing, we introduce a high-throughput rapid method to measure the fitness of different antibiotic-resistant bacterial strains at different selection concentrations. While both the A78C and G70A-G82A double mutant had MIC₉₀ values of at least 2,000 μg/ml, well above all the selection concentrations used, they had vastly different population trajectories at different concentrations of spectinomycin. A78C grew to the highest relative abundance of any strain by the 48- and 72-h time points at the two higher levels of selection, while it was unable to increase in abundance at the lowest level of selection, where G70A-G82A displayed greater fitness (Fig. 3A). Using growth assays with spectinomycin pressure, we find that both doubling time and the length of time it took each mutant strain to exit lag phase likely impacted their success at different antibiotic concentrations (Fig. 3B), with those strains that more rapidly entered logarithmic phase growth and were able to grow more rapidly in the presence of drug having greater success in the competition experiments. This finding is in agreement with Lenski et al., who previously demonstrated that a shortened lag phase plays a key role in fitness and success during bacterial evolution (104). Additionally, some mutants that had comparatively high MIC₉₀ values and were the predominant mutants emerging from the mutagenesis assays, such as G89C, were unable to compete at any level of antibiotic selection. This may be due to the drastic structural change of the G89C mutation, which resulted in an amino acid change of arginine to proline in the S5 ribosomal protein, which could have had detrimental effects on ribosome function. Taken together, the results of this competition experiment suggest that the fitness of individual mutations and their ability to persist in mixed populations is dependent on the concentration of spectinomycin selection.

DISCUSSION

In this work, we define the distinct mutational spectra of several genotoxic stressors in the context of spectinomycin resistance in B. subtilis. Separate studies have reported that stressors such as starvation (29), UV radiation (24), mitomycin C (15), and ROS (105) can generate specific and distinct types of mutations. Specific antibiotic-resistance mutations have been shown to be the result of different types of stress (30–34). We sequenced thousands of individual colonies in order to uncover how genotoxic stress is linked to mutational signature and resulting spectinomycin resistance derived from ribosomal mutations. In doing so, we identified novel spectinomycin-resistance mutations in both the S2 and S5 proteins of B. subtilis and further analyzed individual strains containing base substitutions to characterize the impacts of these mutations on bacterial fitness. Utilizing competition assays, we add to existing evidence that shows the concentration of antibiotic pressure shapes which mutations emerge from a mixed population. This work describes the process by which spectinomycin resistance develops in B. subtilis through a stress-specific pattern of base substitutions, which are narrowed down to a unique spectrum of resistance mutations by a series of selective processes.

Stress-associated mutagenesis leads to spectinomycin resistance through a spectrum of mutations and gradient of selection.

Figure 4 illustrates a process in which B. subtilis exposed to various stressors may undergo mutagenesis that leads to spectinomycin resistance through an initial generation of a stressor-specific spectrum of mutations, followed by selection for viable and resistant mutants. Initially, we show that different forms of stress increase the formation of antibiotic-resistant colonies (Fig. 1A). Next, we show that each stressor is associated with a unique spectrum of mutations (Fig. 1B), in some cases reflecting the DNA-damaging mechanism of the mutagen. This suggests that for certain genotoxic agents, SAM likely plays a greater role in the observed spectrum. Importantly, in this experiment we are only capturing the fraction of the total induced mutations that were selected by two criteria of cell viability followed by spectinomycin resistance. Thus, while the spectrum of all possible mutations is unique for each stressor, the observed mutations are quite limited. In order for a base substitution mutation to be generated, it must be possible in the context of the spectrum of mutation for a given stress condition. For example, mitomycin C-treated colonies often exhibit a G89C mutation that is never observed in the ciprofloxacin-treated colonies, as it is not part of the mutational repertoire of ciprofloxacin. Therefore, development of a resistance through a G89C mutation is only possible given a stress (mitomycin C) that has the potential to generate such a mutation.

FIG 4.

Stress-associated mutagenesis leads to spectinomycin resistance through a spectrum of mutations and a gradient of selection in B. subtilis. A graphical representation of the process by which stress-associated mutagenesis in B. subtilis leads to spectinomycin resistance through the generation of base substitution mutations and subsequent selection. The initial stress causes a stress-specific pattern of mutations but, of all these mutations, most are deleterious or do not impact antibiotic susceptibility (represented by the gray portion of the triangle). Deleterious or growth-inhibiting mutants are lost as they are unable to produce viable cells. Of the viable mutations (denoted by the rainbow in the bottom portion of the pyramid), only a portion of the possible mutations will be able to manifest a resistance phenotype based on fundamental selection criteria, such as that mutation being in the mutational repertoire of the stressor and not impacting cell viability. Transparent cells represent those that were unable to meet the selection requirements for resistance. Finally, the level of antibiotic selection pressure will ultimately determine the final spectrum of base substitution mutations in a mixed population.

There are also clear fundamental barriers to the development of antibiotic resistance through base substitution mutations, such as the ability of mutated cells to tolerate the mutation and proliferate. While our experimental methods only allow us to identify nonlethal mutations under optimal growth conditions, we do illustrate that the A67G mutation in rpsE results in a growth defect at 20°C (Table 1). The ability to grow at given environmental conditions, such as a certain temperature, represents an additional selective hurdle that a mutated cell must cross to become a viable antibiotic-resistant bacterium. Once a bacterium undergoes a mutagenic process that results in viable antibiotic resistance, there is then a potential for competition with other mutants that further narrows the spectrum of antibiotic-resistance mutations (Fig. 4). Resistant bacteria containing different mutations will be more or less competitive depending on the antibiotic concentration present, and thus the encountered antibiotic concentration will further narrow the mutational spectrum (Fig. 3A).

Each genotoxic agent causes a unique spectrum of mutations that reflects mechanisms of DNA damage.

We provide a comprehensive view of spectinomycin resistance base substitution mutations in B. subtilis that allow us to demonstrate the stressor-specific mutational signatures caused by various genotoxic agents. Using a low level of selection (<2× MIC), uncrowded plates, collection of thousands of individual colonies, and next-generation sequencing allowed us identify stressor-specific signatures of mutation. This novel approach allowed us to capture a wide range of base substitutions, including mutations that were both rare and potentially less fit. We have shown that each genotoxic stress causes a unique signature of antibiotic-resistance mutations and that mechanisms of SAM result in mutational spectra that often reflect the means by which the mutagen causes DNA damage (Fig. 1A and B). For example, in stationary-phase cells treated with UV or mitomycin C, the mutational spectrum was highly divergent from untreated stationary-phase cells, suggesting that direct base damage characteristic of these mutagens plays a stronger role in the final mutational spectrum than stationary phase-induced stress (Fig. 1B). While stationary phase stress has been previously shown to be mutagenic (63), we show the additional role that external mutagens play in the mutational spectrum and development of resistance.

DNA maintenance and repair genes contribute to the frequency and spectrum of stress-associated spectinomycin-resistance mutations in B. subtilis.

Using a set of B. subtilis gene knockouts, each deficient in a single gene involved in DNA maintenance or repair, we show that the presence of certain genes can limit or exacerbate the generation of antibiotic resistance through mitomycin C-induced mutagenesis (Fig. 2A). Our data show that the radA, sbcD, yobH, yozK, uvrX, exoAA, mfd, and dinB genes are essential to the mutagenic activity of mitomycin C in B. subtilis. Interestingly, some of these genes are involved in the SOS response of B. subtilis while others are not, suggesting that mutagenesis may be dependent on the activity of multiple DNA repair pathways (91, 92). On the other hand, strains deficient in adaA, yhaO, or mutSB had an increased frequency of spectinomycin-resistant colony formation when exposed to mitomycin C. Understanding which DNA repair genes are involved in altering the rate of mutagenesis could be valuable in limiting the development of antibiotic resistance.

Impact of a spectinomycin selection gradient on the fitness of resistant strains.

The development of antibiotic resistance is highly complex and involves a myriad of factors. The concentration of the antibiotic selection is one such factor that plays a key role in determining the evolutionary trajectory of a bacterial population when selecting for resistance (106). While using a fixed selection concentration allowed us to understand the spectrum of mutations caused by different mutagens, this homogeneous selective force is not representative of antibiotic concentrations in nature and is often a constraint on studies of the evolution of resistance (107). In therapeutic and naturally occurring settings, antibiotics are part of heterogeneous environments, resulting in concentration gradients (108, 109). It is vital to understand the role of these gradients in the selection of antibiotic-resistance mutations. Here, we build on the idea of “selective compartments,” put forth by Baquero and Negri (107), by studying the selection of different mutations across a gradient of antibiotic concentrations and by assessing mutational spectra and mutant fitness in the context of no, low, medium, and high antibiotic selection concentrations. In competition experiments using spectinomycin concentrations varying by less than one order of magnitude, we obtained highly disparate final populations with certain mutations showing a distinct fitness advantage at different drug concentrations (Fig. 3A). Furthermore, these results show that the strength of antibiotic selection modulates the fitness of individual mutations based not on MIC alone. Instead, the fitness of mutations under selection is likely determined by growth dynamics, namely, growth rate and the length of time to exit lag phase, as well as by MIC (Fig. 3A and B, Table S3).

Limitations.

This work has several limitations intrinsic to the system and methodologies that were used. While we were able to identify unique mutation spectrums for various bacterial stressors, this study only looked at the mutations from a single concentration of a handful of stressors at a set time point. Different lengths of exposure could significantly alter the mutational spectrum of each stressor. In addition, our exposures were selected to limit the number of bacterial division cycles to approximately one division, and thus we would not detect additive mutations acquired through cycles of mutation and replication. While we could infer the mechanism behind mutagenesis for stressors that exhibited a signature mutation, we do not provide direct evidence for the molecular basis of DNA damage associated with each stressor, although, in many cases, these mechanisms have been thoroughly validated in the literature. Additionally, we mainly focus on the spectrum of mutations in a single gene, rpsE, in which resistance develops primarily through base substitution mutations, excluding the potential for studying insertion, deletion, or frameshift mutations. Therefore, we are unable to capture the “full” mutational spectrum, which also includes regions outside of rpsE as well as any deleterious mutations that did not make it through the selection process. Further work is necessary to fully define the array of deleterious mutations to understand the mutational spectrum of each stressor tested here, which would require whole-genome sequencing of both viable and unviable cells. Furthermore, we utilized the model organism B. subtilis, but it is possible that these stressors exhibit different spectra of mutation in other organisms based on taxon-specific stress responses. Finally, in terms of selection, we examined a fraction of the selection factors that shape which mutations are able to manifest viable, antibiotic-resistant bacteria. There are many more drug concentrations, growth conditions, and other selective forces that are important for the development of antibiotic resistance in a population. Thus, future studies are needed to explore the role of other stressors and selective factors in shaping the development of antibiotic resistance.

MATERIAL AND METHODS

Bacterial strains.

The wild-type strain of Bacillus subtilis was B. subtilis 168, obtained from the Bacillus Genetic Stock Center (Columbus, OH, USA) (http://www.bgsc.org/index.php). All spectinomycin-resistant mutant strains were derived from this strain with the exception of those resistant isolates derived from the DNA-repair gene knockout strains in Fig. 2. The DNA-repair knockout strains were originally generated by Koo et al. (90) and were obtained from the Bacillus Genetic Stock Center (Columbus, OH, USA) (http://www.bgsc.org/index.php).

Stress conditions.

Growth for all experiments occurred in lysogeny broth (LB) at 37°C unless otherwise specified. Doses of each stressor were determined by dose titration experiments, with the goal of inducing mutagenesis and minimizing cell mortality or growth inhibition. The following concentrations/doses of each drug were used: 100 ng/ml mitomycin C, 600 ng/ml ciprofloxacin, 118.5 μg/ml CoCl₂, 0.3125 mM hydrogen peroxide, and 500 J/m² UV, administered using an FB-UVXL-1000 UV crosslinker from the Spectronics Corporation (Westbury, New York).

Spectinomycin mutant generation experiments.

(i) Logarithmic-phase growth. During logarithmic-phase growth experiments, cells were grown overnight in SpC medium (Table S7) for ∼16 h to an OD₆₀₀ of at least 0.8. From the overnight cultures, subcultures were seeded 1:333 (WT strain) or 1:143 (Δ strains) into LB. Cultures were grown at 37°C with shaking at 250 rpm, to an OD₆₀₀ of ∼0.5 and stressors were then administered. Cells were grown with stressors for 30 min at 37°C and shaking at 300 rpm; exceptions were for UV treatments involving a short exposure followed by 30 min of growth, and CoCl₂, which was administered for 60 min. Following the exposure to stressors, cells were washed with LB and grown in an equal volume of LB for 30 min at 37°C, with shaking at 300 rpm. Following outgrowth, 40 μl of cells was serially diluted 10-fold in phosphate-buffered saline (PBS) and plated onto LB agar (without selection) for counting of CFU. The remainder of the cells was bead spread on LB plates containing 100 μg/ml spectinomycin. Plates were grown for 16 h at 37°C, at which point CFU counts were taken from LB plates. Selection plates were grown for an additional 24 h at room temperature and colonies were counted.

(ii) Stationary-phase growth. For the stationary-phase growth experiments, cells were grown overnight in SpC medium (Table S7) for ∼16 h to an OD₆₀₀ of at least 0.8. From the overnight cultures, subcultures were seeded 3:7 (WT strain) into LB. Cultures were grown at 37°C with shaking at 250 rpm to an OD₆₀₀ of ∼1.8, at which point stressors were administered. Cells were exposed to stressors for 3 h at 37°C with shaking at 300 rpm, except for UV treatments, which involved a short exposure followed by 3 h of incubation. Following the exposure to stressors, cells were washed with LB and incubated in an equal volume of LB for 30 min at 37°C with shaking at 300 rpm. Following outgrowth, 40 μl of cells was serially diluted 10-fold in PBS and plated on LB agar without selection for counting CFU. The remainder of the cells was bead spread on LB plates containing 100 μg/ml spectinomycin. Plates were grown for 16 h at 37°C, at which point CFU counts were taken from LB plates. Selection plates were grown for an additional 24 h at room temperature and colonies were counted.

Isolation, growth, and DNA extraction of spectinomycin-resistant mutants.

Individual colonies were manually isolated from LB spectinomycin selection plates using pipette tips. Each colony was grown in individual wells of 96-well plates containing 200 μl of LB medium with 100 μg/ml spectinomycin. Plates were grown for 16 h at 37°C with shaking at 250 rpm and adjusted to have equal cell density in each well. Then, 100 μl from each well was taken and pooled for DNA extraction. DNA extraction was performed using the ZymoBIOMICS DNA miniprep kit from Zymo Research (Irvine, CA, USA) following the manufacturer’s instructions, with final elution in 100 μl of molecular grade H₂O. Total DNA concentration was measured using the SpectraMax M3 microplate reader from Molecular Devices LLC (San Jose, CA, USA) using the DNA Quantitation with the SpectraDrop Micro-Volume Microplate protocol.

Illumina sequencing amplicon library preparation and sequencing.

The rpsE gene was amplified from DNA derived from ∼500 colonies from each condition. The primers used for generating amplicon libraries are listed in Table S5. A 300-bp region of the rpsE gene was amplified using a common reverse primer and barcoded forward primer (Table S5), with primers for each condition using a unique barcode. PCRs for each condition were performed in 25 μl triplicate reactions in a T100 thermal cycler from Bio-Rad (Hercules, CA, USA) under the following conditions: 180 s at 98°C, 45 s at 98°C, 60 s at 60°C, 90 s at 72°C, steps 2 to 4 repeated 34×, and a final step of 600 s at 72°C. PCR products of the triplet reactions of each condition were combined and purified using the NucleoSpin Gel and PCR clean-up kit from Macherey-Nagel Inc. (Düren, Germany). Purified amplicons (240 ng) from each condition were pooled together for the final sequencing library, which was sequenced at the Rhode Island Genomics and Sequencing Center at the University of Rhode Island (Kingston, RI, USA). Amplicons were paired-end sequenced (2 × 250 bp) on an Illumina MiSeq platform using a 600-cycle kit with standard protocols.

Illumina amplicon sequencing analysis.

Raw paired-end FASTQ files were demultiplexed using idemp (https://github.com/yhwu/idemp). Reads were quality filtered, trimmed, de-noised, and merged using DADA2 (110). Representative sequences (unique mutations) were determined and extracted using QIIME2 (version 2020.8) (111). Processed reads were matched to representative sequences using the QIIME2 “Closed-reference clustering” tool with vsearch (112) and the percent identity set to 1.00 (–p-perc-identity 1.00) to only identify exact matches to representative sequences. Protein models of the mutations were generated using VMD version 1.9.3 (113).

Generation of spectinomycin-resistant base substitution mutant strains.

To generate single-nucleotide rpsE mutant strains, PCR-amplified DNA of individually isolated spectinomycin-resistant colonies was used for natural transformation into the WT Bacillus subtilis 168 background strain. Natural transformation allowed us to transfer only the region of DNA containing the select S2 base substitution mutations we wanted to test, thus excluding any other possible sites of mutation that might have occurred in the generation of the initial spectinomycin-resistant colonies. DNA from 11 different colonies that were previously Sanger sequenced was used to amplify the rpsE. The same protocol used for generating the Sanger sequencing amplicons was used to generate the DNA used for transformation into the WT strain. Details of this protocol can be found in the “Sanger sequencing” section of these Materials and Methods. The purified rpsE amplicons were then incorporated into the WT background using the following procedure for natural transformation. First, WT B. subtilis 168 was grown overnight (∼16 h) in LB at 37°C with shaking at 300 rpm. Second, the overnight culture was diluted 1:100 into 5 ml of freshly prepared MNGE medium (Supplemental Table S7) and grown for 7 h at 37°C with shaking at 250 rpm, to an OD₆₀₀ of ∼0.5. Third, 300 to 500 ng of the amplified rpsE gene from the previous PCR step was added to 400-μl aliquots of the cells, which were then grown for 60 min at 37°C with shaking at 300 rpm. Fourth, 100 μl of expression mix was added to each 400-μl culture and grown for an additional 60 min at 37°C with shaking at 300 rpm. Finally, the cells from each culture were pelleted, the supernatant was removed, and the pellet was resuspended in 200 μl of LB and spread onto LB plates containing 100 μg/ml spectinomycin using glass beads. Plates were grown at 37°C for 24 h and individual colonies were picked and grown in 2 ml of LB broth containing 100 μg/ml of spectinomycin for selection. Frozen stocks of these mutants were generated and a portion of the cells was used to perform Sanger sequencing as described in the “Sanger sequencing” section of these Materials and Methods to confirm the mutations present in the strains in Table 1 and Fig. 3. Upon sequencing the isolates to confirm mutations, we identified 7 isolates with the expected mutation and 4 containing mutations (A78C, G83A, G83A-G151T, and G70A-G82A) that developed spontaneously and did not correspond to the DNA added to the natural transformation assay.

Competition experiments between spectinomycin-resistant base substitution mutant strains.

The strains used for the competition experiment were generated as described above. Using the 11 mutants listed in Table 1, as well as WT B. subtilis 168, we performed growth competition experiments with no antibiotic selection or in medium containing 62.5 μg/ml, 250 μg/ml, or 1,000 μg/ml of spectinomycin. These drug concentrations correspond to the MIC of the WT strain (62.5 μg/ml), the lowest MIC of the 11 isolates tested (250 μg/ml), and the second highest MIC of all the isolates (1,000 μg/ml). For each of the conditions, there were four biological replicates. Four cultures of each of the 12 strains used for this competition were grown overnight (∼16 h) at 37°C, with shaking at 300 rpm in LB broth with 100 μg/ml antibiotic added to the 11 spectinomycin-resistant isolates. No selection was added to the WT cultures. The OD₆₀₀ of overnight cultures was taken and all 12 strains were added in equal proportions into a new culture containing LB broth with 0, 62.5, 250, or 1,000 μg/ml spectinomycin, with a final combined OD₆₀₀ of 0.1. A 2-ml sample of cells was taken from the initial no-antibiotic replicates immediately before starting incubation and was used for the T:0 time point for all concentrations. Cultures were then grown at 37°C, shaking at 250 rpm, and 500 μl of cells was taken at 3, 6, 12, and 24 h. Immediately after the 24-h time point, the cultures were passaged into fresh medium (maintaining antibiotic selection) to an OD₆₀₀ of 0.01. Additional samples were taken from cultures at 48 h, immediately followed by another passaging of cells, and at 72 h after the initial T:0 time point. DNA from cells collected at each time point was extracted using the ZymoBIOMICS DNA miniprep kit from Zymo Research (Irvine, CA, USA) immediately after sampling. The extracted DNA was quantified using the Qubit High Sensitivity reagent with a Qubit 3.0 fluorometer from Thermo Fisher Scientific (Waltham, MA, USA). The 24-h, 250 μg/ml, replicate D sample had an issue with the extraction and could not be used for downstream processing.

Nanopore MinION sequencing amplicon library preparation and sequencing.

The samples from the competition experiment were prepared for sequencing on the Oxford Nanopore MinION (Oxford Nanopore Technologies, Oxford, UK). Due to the limitation of having only 96 barcodes and 100 samples, the 72-h, no-antibiotic samples were not sequenced. The rpsE gene was amplified from the DNA of each sample taken during the competition experiment. PCRs for each sample were performed in 25-μl triplicate reactions in a T100 thermal cycler from Bio-Rad (Hercules, CA, USA) under the following conditions: 180 s at 98°C, 45 s at 98°C, 60 s at 60°C, 90 s at 72°C, steps 2 to 4 repeated 34×, and a final step of 600 s at 72°C. PCR products of the triplicate reactions of each condition were combined and purified using the NucleoSpin Gel and PCR clean-up kit from Macherey-Nagel Inc. (Düren, Germany) and DNA was quantified using a Qubit 3.0 fluorometer from Thermo Fisher Scientific (Waltham, MA, USA).

Cleaned, quantified amplicons were then prepared for sequencing using the Native Barcoding Expansion 96 (EXP-NBD196) and Ligation Sequencing kit (SQK-LSK109) from Oxford Nanopore Technologies (Oxford, UK). The protocol was followed according to the manufacturer’s instructions with any modifications detailed here. The protocol began with 240 fmol (50 ng of 320-bp amplicon) of DNA for each reaction. End prep was performed using the NEBNext Ultra II End repair/dA-tailing module (E7546) from New England BioLabs (NEB) (Ipswich, MA, USA) and the end repair reaction was performed in a T100 thermal cycler from Bio-Rad (Hercules, CA, USA) at 20°C for 20 min and 65°C for 20 min. The native barcode ligation was performed using the Native Barcoding Expansion 96 (EXP-NBD196) (Oxford Nanopore Technologies, Oxford, UK) in conjunction with NEB Blunt/TA Ligase Master Mix (M0367) from New England BioLabs (Ipswich, MA, USA) according to the manufacturer’s instructions. The barcoded DNA was quantified using the Qubit High Sensitivity reagent with a Qubit 3.0 fluorometer from Thermo Fisher Scientific (Waltham, MA, USA). Adapter ligation was performed using the Ligation Sequencing kit (SQK-LSK109) from Oxford Nanopore Technologies (Oxford Nanopore Technologies, Oxford, UK) and NEBNext Quick Ligation module (E6056) from New England BioLabs (Ipswich, MA, USA) according to the manufacturer’s instructions. The final library was quantified using the Qubit High Sensitivity reagent with a Qubit 3.0 fluorometer from Thermo Fisher Scientific (Waltham, MA, USA). Priming and loading the SpotON Flow Cell was performed using the Flow Cell Priming kit (EXP-FLP002) (Oxford Nanopore Technologies, Oxford, UK) according to the manufacturer’s instructions. We performed sequencing on the Oxford Nanopore MinION using MinKNOW (MinION Mk1B software) (Oxford Nanopore Technologies, Oxford, UK) with the default protocol and live basecalling turned OFF. The run was terminated after ∼48 h, as none of the pores had shown sequencing activity for several hours.

Nanopore MinION sequencing analysis.

Raw sequences were processed using the Guppy (version 4.0.11) software (Oxford Nanopore Technologies, Oxford, UK) via the command line interface. Basecalling was performed using the guppy_basecaller command with the default settings including use of the high accuracy (HAC) model. Basecalled sequences were then demultiplexed using the guppy_barcoder command with default settings. Demultiplexed reads were then filtered using the Filtlong command with the following parameters: –min_length 400 –min_mean_q 9 –trim. Filtered reads were then mapped to a reference of the rpsE sequences from the 12 strains used in the competition experiment using the minimap2 version 2.17 mapping software (114). Finally, only mapped reads with 100% identity to the reference sequences were used for downstream analyses.

Statistical methods for competition experiments.

Generalized estimating equations were used for all hypothesis testing. Observations drawn from within a particular tube (and its subsequent passage tube) were nested as having correlated residual error. Relative abundance was modeled as binomial, with each strain’s count per total count across all strains. Adjusted counts were modeled as Poisson with the natural log of the absolute amount of starting DNA added to the library preparation as an offset in the model (creating dependent variable units as counts per ng of DNA input into the Nanopore library prep). All models also implemented classical sandwich estimation to adjust for how empirical variances may have differed from model assumptions. Comparisons were made between all time points for the no-drug selection group, and all time points except T:0 for the groups in which spectinomycin was added. A total of 163 hypothesis tests for relative changes between aliquots were carried out as orthogonal linear estimates (4 means) for each strain, maintaining an alpha of 0.05 using the Holm test to adjust each P value. The above statistical analyses were performed using Statistical Analysis Software, SAS (Cary, NC, USA).

Whole-genome sequencing.

Four spectinomycin-resistant isolates with MICs of >1,000 μg/ml were selected for whole-genome sequencing. DNA from these selected isolates was extracted using the ZymoBIOMICS DNA miniprep kit from Zymo Research (Irvine, CA, USA) and quantified using a Qubit 3.0 fluorometer from Thermo Fisher Scientific (Waltham, MA, USA). The metagenomic library was prepared using the NEB Next Ultra II DNA Library prep kit from New England BioLabs (Ipswich, MA). Metagenomic libraries were sequenced on a NovaSeq 6000 or ISeq100 instrument (San Diego, CA).

Whole-genome sequencing analysis.

Reads were trimmed with Trimmomatic (version 0.36) with SLIDINGWINDOW set at 4:20, MINLEN set at 50, and ILLUMINACLIP: TruSeq3-PE.fa:2:20:10 (115). Trimmed reads were assembled and single nucleotide variants were searched for using the Variation Analysis tool with BWA-mem/FreeBayes setting of the PATRIC webserver (version 3.5.38) (116).

Sanger sequencing.

A portion of individual colonies used for Sanger sequencing of the rpsE and rpsB genes were boiled at 100°C for 15 min in 20 μl of Tris-EDTA (TE) buffer to lyse cells. A 1-μl aliquot of a 1:10 dilution of boiled cell lysate was then used as the template for PCR of the genes of interest. Primers pairs 388/389 and 553/554 were used to amplify the rpsE and rpsB genes, respectively (Table S4). PCRs for each gene were performed in 25-μl reactions in a T100 thermal cycler from Bio-Rad (Hercules, CA, USA) under the following conditions: 180 s at 98°C, 45 s at 98°C, 60 s at 55°C, 90 s at 72°C, steps 2 to 4 repeated 34×, and a final step of 600 s at 72°C. PCR products were then purified using the NucleoSpin Gel and PCR clean-up kit from Macherey-Nagel Inc. (Düren, Germany) and DNA was quantified on the SpectraMax M3 microplate reader from Molecular Devices LLC (San Jose, CA, USA) using the DNA Quantitation with the SpectraDrop Micro-Volume microplate protocol. Amplicons were sequenced at Eurofins Genomics. Alignments of the Sanger sequencing data were performed using Unipro UGENE (version 1.32.0) (117).

MIC determination.

MICs were determined using the broth dilution method (118). Bacillus subtilis strains were grown overnight in LB. Overnight cultures were diluted 1:10,000 and added to a 96-well plate. Spectinomycin was added to a concentration of 1 mg/ml to cell culture medium and serially diluted 2-fold across the plate. Cells were then incubated at 37°C with shaking at 300 rpm for ∼20 h and the OD₆₀₀ was taken to measure growth using the SpectraMax M3 microplate reader from Molecular Devices LLC (San Jose, CA, USA). MIC₉₀ was recorded for all MIC experiments.

Growth rate determination and time to exit lag phase.

Overnight cultures of B. subtilis strains grown in LB were diluted to an OD₆₀₀ of ∼0.05 and then grown in LB in triplicate at 37°C with shaking at 300 rpm for 390 min for no-drug conditions or 510 min with spectinomycin present, with OD₆₀₀ measurements taken every 30 min using the SpectraMax M3 microplate reader from Molecular Devices LLC (San Jose, CA, USA). The growth assays were performed with 0, 62.5, 250, and 1,000 μg/ml spectinomycin with 6 to 12 replicates in each group. To determine the doubling time of each strain, the growth curves from OD₆₀₀ ∼0.15 to 0.4 were fitted to an exponential growth function using default settings in Prism (version 8.0). The time to exit lag phase was determined by fitting a simple linear regression to estimate the time at which cultures reached OD₆₀₀ = 0.2. An ANOVA with an alpha of 0.05 was used to compare the doubling time and time to exit lag phase for each strain.

Cold sensitivity determination.

Selected spectinomycin-resistant strains were grown in LB with spectinomycin (100 μg/ml) in triplicate. Cultures were serially diluted 10-fold in PBS and 5 μl of each dilution was spot plated on LB plates with two sets of plates for each replicate. One set of replicates was grown for 18 h at 37°C and the other for 84 h at 20°C.

Data availability.

Raw Illumina rpsE amplicon sequencing reads were deposited in the NCBI Sequence Read Archive under the BioProject number PRJNA703389. The base-called Nanopore rpsE amplicon sequencing reads were deposited in the NCBI Sequence Read Archive under the submission number PRJNA704934. Raw Illumina whole-genome sequencing reads from the 12 strains used for the competition experiment were deposited in the NCBI Sequence Read Archive under the submission PRJNA748029.