# Ribosomal mutations promote the evolution of antibiotic resistance in a multidrug environment

1. The Broad Institute of MIT and Harvard, United States
2. Massachusetts General Hospital, United States
3. Harvard Medical School, United States
4. Texas A&M University, United States
5. Harvard T.H. Chan School of Public Health, United States
Research Article

## Abstract

Antibiotic resistance arising via chromosomal mutations is typically specific to a particular antibiotic or class of antibiotics. We have identified mutations in genes encoding ribosomal components in Mycobacterium smegmatis that confer resistance to several structurally and mechanistically unrelated classes of antibiotics and enhance survival following heat shock and membrane stress. These mutations affect ribosome assembly and cause large-scale transcriptomic and proteomic changes, including the downregulation of the catalase KatG, an activating enzyme required for isoniazid sensitivity, and upregulation of WhiB7, a transcription factor involved in innate antibiotic resistance. Importantly, while these ribosomal mutations have a fitness cost in antibiotic-free medium, in a multidrug environment they promote the evolution of high-level, target-based resistance. Further, suppressor mutations can then be easily acquired to restore wild-type growth. Thus, ribosomal mutations can serve as stepping-stones in an evolutionary path leading to the emergence of high-level, multidrug resistance.

https://doi.org/10.7554/eLife.20420.001

## eLife digest

The rise of antibiotic resistant bacteria is challenging clinicians, and some infections are now resistant to almost all of the drugs that are currently available. Some types of bacteria – such as mycobacteria, which include the bacteria that cause tuberculosis and leprosy – can only acquire antibiotic resistance from mutations that alter their existing genes. The process by which bacteria develop resistance to multiple drugs is generally viewed as a stepwise accumulation of different mutations. However, the role of individual mutations that increase a bacterium’s resistance to multiple antibiotics has not been fully explored.

Gomez, Kaufmann-Malaga et al. exposed bacteria from the species Mycobacterium smegmatis, a cousin of the bacterium that causes tuberculosis, to a mixture of relatively low concentrations of different antibiotics that should kill the bacteria relatively slowly. Hundreds of small bacteria cultures were grown in parallel, and only a fraction of them developed antibiotic-resistant members. Gomez, Kaufmann-Malaga et al. identified mutations in these bacteria that unexpectedly gave the bacteria resistance to several unrelated classes of antibiotics.

Individual mutants carried single mutations in different components of the ribosome, a complex molecular machine that helps to build proteins inside cells. As well as increasing their resistance to antibiotics, these mutations also reduced the growth rate of the bacteria. This meant that when the bacteria were grown in an antibiotic-free environment they survived less well than non-mutant bacteria. However, the mutations gave the bacteria an advantage in environments that contained many different antibiotics, as they could more easily develop mutations that made them more resistant to other drugs. Thus, the mutant bacteria can serve as stepping-stones toward the development of high-level resistance to multiple drugs.

Further work will now explore whether this phenomenon occurs in a range of other bacterial species, including the bacteria that cause tuberculosis. While new antibiotics are desperately needed, a better understanding of how bacteria evolve the ability to resist the effects of antibiotics will help us to preserve the usefulness of existing and future drugs.

https://doi.org/10.7554/eLife.20420.002

## Introduction

Antibiotic resistance is on the rise in virtually all clinically relevant pathogens. A 2013 report from the CDC (Centers for Disease Control and Prevention (CDC), 2013) identified the most significant antibiotic resistant pathogens, including S. pneumoniae, extended-spectrum β-lactamase producing Enterobacteriaceae, vancomycin-resistant Enterococci, and methicillin-resistant Staphylococcus aureus (MRSA), which has now achieved pandemic spread (Monecke et al., 2011). This report estimates that in the United States, antibiotic resistant infections kill at least 23,000 people each year. The threat of antibiotic resistant bacteria is however not limited to the developing world, as the prevalence of multidrug resistant (MDR) and extensively drug resistant (XDR) tuberculosis (TB) has been on the rise for the last decade (Migliori et al., 2010), and untreatable gonorrhoea infections are now emerging (Unemo and Nicholas, 2012). A U.K. task force estimates up to 10 million deaths each year due to antibiotic resistant infections world-wide by 2050 (O’Neill, 2016). A better understanding of the events leading to the generation and fixation of resistance-conferring mutations is vital to preserving the effectiveness of current and future antibiotics.

Antibiotic resistance can be acquired by horizontal gene transfer or mutation of existing genes. Mobile elements often carry genes encoding inactivating enzymes (Ramirez and Tolmasky, 2010; Bush, 2013), efflux systems (Poole, 2005), or alternative enzymes that bypass the native enzyme targets of the antibiotic, as is seen with MRSA and vancomycin resistant Enterococcus (de Lencastre et al., 1994; Courvalin, 2006; Wellington et al., 2013). However, some bacteria do not readily acquire new DNA, including pathogenic mycobacteria (Musser, 1995). Resistance can nonetheless emerge through acquisition of chromosomal mutations that confer resistance in a variety of ways, including altering the target to prevent antibiotic binding (Musser, 1995; Jacoby, 2005), increasing target expression (Banerjee et al., 1994; Rouse et al., 1995), decreasing intracellular drug concentration via enhanced efflux or reduced permeability (Fernández and Hancock, 2012), or reducing the activation of prodrugs (Scorpio and Zhang, 1996; Zhang et al., 1992). With these many mutational pathways to resistance available, it is imperative to understand the factors that contribute to the de novo development of antibiotic resistance.

Clinically, the emergence of antibiotic resistance via mutation can be reduced by ensuring that tissue concentrations of antibiotic always exceed a threshold known as the mutant prevention concentration (MPC) (Zhao and Drlica, 2001; Martinez et al., 2012; Drusano, 2004; Baquero and Negri, 1997). Above this level, no single mutation can decrease antibiotic sensitivity sufficiently to allow bacterial growth or survival. However, in patients, such concentrations can be difficult to achieve and maintain due to pharmacokinetic and toxicity issues. Concentrations below the MPC but above the minimum inhibitory concentration (MIC) define the traditional mutant selection window (Drlica, 2003), the range of concentrations in which a single mutation can confer a selective advantage. However, even concentrations well below the MIC can select for resistant organisms (Gullberg et al., 2011; Liu et al., 2011), and the enhanced fitness of even low-level resistant mutants can contribute to the development of high-level resistance (Baquero et al., 1998).

The use of antibiotics in combination can also limit the emergence of resistance (Mouton, 1999) by requiring a bacterium to acquire multiple mutations simultaneously in order to survive in a multidrug environment (Lipsitch and Levin, 1997; Fischbach, 2011). However, multi-antibiotic therapy can be undermined by issues including poor compliance, low quality antibiotics, and inadequate susceptibility data (Ormerod, 2005). Combinations of antibiotics with different pharmacokinetic properties, along with pharmacogenomic differences between individuals, can also result in periods where one or more of the antibiotics is present at subinhibitory concentrations, wherein a single mutation could allow for an enrichment of a monoresistant strain, which would in turn increase the likelihood of the emergence of a strain carrying multiple, independent resistance-conferring mutations (Mitchison, 1998; Ramachandran and Swaminathan, 2012). Also, in many infections, highly dynamic bacterial populations can reach high numbers (>1010 organisms) and thus the likelihood of a bacterial lineage becoming MDR during the course of a single human infection may be more common than had been previously perceived (Pasipanodya and Gumbo, 2011; Colijn et al., 2011).

Antibiotic stress itself can also impact rates of mutation. Increased mutational frequencies can be a direct result of antibiotic action, as seen with the DNA damaging effects of fluoroquinolones (Cirz et al., 2005; Cirz and Romesberg, 2006), or due to downstream effects of antibiotics, which have been suggested to involve free radical production resulting in damage to the bacterial chromosome (Kohanski et al., 2010; Foti et al., 2012; Kohanski et al., 2007; Dwyer et al., 2014). While DNA damage may contribute to the bactericidal activity of antibiotics, it may also promote resistance-conferring mutations. Recently, several investigators have shown mutation rates can be elevated even by exposure to subinhibitory levels of antibiotic (Monecke et al., 2011; Kohanski et al., 2010; Blázquez et al., 2012; Andersson and Hughes, 2012).

To gain insight into the molecular processes that promote antibiotic resistance, we investigated the relationships between antibiotic concentration, bactericidal activity, and the emergence of resistant mutants in a model species in which resistance arises nearly exclusively through mutation, Mycobacterium smegmatis. We found novel ribosomal mutations that confer resistance to antibiotics of several different classes that do not target the ribosome, as well as to non-antibiotic stressors. Despite the fact that these mutations have an associated fitness cost in the absence of antibiotic stress and confer relatively modest levels of resistance, they can facilitate evolution to high levels of antibiotic resistance while escaping the fitness cost associated with these mutations through a subsequent reversion. Such transient mutations can thus serve as stepping-stones that facilitate evolution from antibiotic sensitivity to multidrug resistance.

## Results

### Selection for low level ciprofloxacin resistance identifies ribosomal mutants

We developed a microtiter-based liquid culture system that enables the isolation of individual antibiotic-resistant mutants while simultaneously monitoring the growth and kill kinetics of the bulk, susceptible population (Figure 1A). We uniformly seeded 384-well plates with green fluorescent protein (GFP)-expressing M. smegmatis (10 bacilli per well), allowed the bacterial populations in each well to expand in parallel for several generations in the manner of Luria and Delbruck (Luria and Delbrück, 1943), and then exposed them to antibiotics. At the start of antibiotic exposure, the per-well population was sufficiently small so that fluorescent signal in each well was below the threshold of detection, which is approximately 105 cells/well (Figure 1—figure supplement 1). Over time, fluorescent signal appeared in a fraction of wells due to the outgrowth of an antibiotic-resistant mutant. At regular intervals during antibiotic exposure, eight wells were sacrificed to assess bacterial viability using a most probable number (MPN) assay in order to measure the kinetics of bacterial killing in response to antibiotic treatment (Figure 1—figure supplement 1). This sampling showed an initial expansion of the bacterial population to approximately 300 cells per well by 12 hr, followed by a rapid decline during the next 36 hr.

Figure 1 with 2 supplements see all

The relatively slow rate of killing by fluoroquinolone (FQ) antibiotics at levels just above the minimum inhibitory concentration (MIC) allows for a prolonged survival time under antibiotic stress, providing an enhanced opportunity for the acquisition of mutations. When we exposed M. smegmatis to the FQ ciprofloxacin (CIP) at 0.3 μg/ml (1.2X MIC), we observed the emergence of fluorescent signal in 19 of 768 wells, monitoring fluorescence at daily intervals (Figure 1B). This fluorescence was due to the outgrowth of CIP resistant M. smegmatis; we have confirmed that, in this system, the lack of signal within a well is due to the CIP-dependent killing of CIP-sensitive bacteria and not to plasmid loss in the mc2155 strain, which has been shown to have altered plasmid maintenance (Panas et al., 2014) (Figure 1—figure supplement 2). This yields a mutation frequency of 9.7 × 10?5 based on the peak population of approximately 105 bacteria per 384 well plate seen at 12 hr after introduction of CIP (Figure 1—figure supplement 1B), prior to the onset of killing. We measured resistance levels of clones cultured from each of the 19 mutants (as IC50, see Materials and methods), and found IC50 increases ranging from a 2.2 to 44-fold relative to wild-type (Figure 1C).

The target of FQs is the A subunit of DNA gyrase, encoded by gyrA. Mutations here occur at a frequency of approximately 10?7 and cause a 12–20 fold increase in MIC to CIP (Zhou et al., 2000). We thus PCR amplified and sequenced gyrA from each of the 19 isolated mutants, but none had mutations at this locus. We then performed whole genome sequencing (WGS, Table 1) of each mutant, which revealed resistance-conferring mutations in the lfrR gene in four mutants (21%). The lfrR gene encodes a transcriptional repressor of the LfrA efflux pump (Buroni et al., 2006). LfrA contributes to the intrinsic resistance of M. smegmatis to FQs (Takiff et al., 1996; Sander et al., 2000), and de-repression of lfrA was shown to increase the MIC to CIP 16-fold (Drusano, 2004). The isolated lfrR mutants were significantly more resistant to CIP (mean 34-fold increase in IC50 over wild-type, Figure 1C), than the other 15 mutants (mean 7-fold increase, p=0.0003). The lfrR mutants also arose more rapidly, with 3/4 reaching detectable fluorescence levels within 3 days. In contrast, the 15 non-gyrA, non-lfrR mutants arose more slowly, taking up to 7 days to reach detectable fluorescence (Figure 1B).

Table 1

When we examined the genome sequence of the 15 non-gyrA, non-lfrR mutants, 14/15 had mutations within genes associated with the ribosome. Ten contained small insertions or deletions in operons encoding ribosomal proteins (Figure 2A, Table 1), including seven with frameshift mutations in proteins composing the large ribosomal subunit: 5 in rplO (L15), 1 in rplF (L6), and 1 in rplY (L25). One mutant had an intergenic deletion between rplW and rplB (L23/L2), and two had frameshifts in rpsE, encoding the S5 protein of the small subunit. Four of the five remaining mutants had SNPs in rrlB (23S rRNA), one of two ribosomal RNA operons in M. smegmatis (Sander et al., 1996). We were unable to identify any causal mutations in the final mutant using WGS. We confirmed that these ribosomal mutations were directly responsible for FQ-resistant phenotype by engineering selected mutations (rplO-1 and rpsE-1; see Table 1) into the chromosome of M. smegmatis using allelic exchange (Figure 2B) and measuring the corresponding MICs to CIP. Complementation by a plasmid expressing a wild-type rplO allele restored CIP sensitivity to the rplO-1 mutant (Figure 2—figure supplement 1).

Figure 2 with 1 supplement see all

### Ribosomal mutants are resistant to multiple antibiotics and other stresses

We tested sensitivity to nine additional antimicrobials, hypothesizing that mutations in the translational machinery might have broad effects on innate antibiotic sensitivity given their unexpected impact on FQ sensitivity. Intriguingly, ribosomal mutants showed cross-resistance to antibiotics of unrelated classes to which they had never been exposed (Figure 2C), as well as collateral sensitivity to meropenem. The most pronounced shifts were to the cell-wall biosynthesis inhibitor isoniazid (INH), with an average 40-fold increase in IC50 over wild-type (range 6.5 to 158-fold). In contrast, lfrR mutants showed little cross-resistance to INH, consistent with previous characterizations of this efflux system. Cross-resistances were not due to acquisition of additional mutations during the MIC assay, because the assay duration does not allow single cells, i.e. de novo mutants, to reach detectable levels of fluorescence. The rplO-1 and rpsE-1 allelic exchange mutants also displayed these phenotypes (Figure 2—figure supplement 1). We also evaluated the ability of ribosomal mutants to tolerate non-antibiotic stresses. Ribosomal mutants were better able to tolerate membrane stress due to SDS exposure (Figure 2D) and high temperature (54°C, Figure 2E) compared to wild-type bacteria. Altogether, these data demonstrate that perturbations to the ribosome can have broad ranging consequences on bacterial survival under duress, leading us to investigate the mechanisms underlying these phenotypes.

The late emergence of ribosomal mutants during CIP selection (Figure 1A) suggested that they might have growth defects. We thus first measured doubling times of the ribosomal mutants both in the presence and absence of CIP (Figure 3A). In contrast to the 2.4 hr doubling time calculated for the parental wild-type strain or lfrR mutants, ribosomal mutants had doubling times ranging from 4–6.5 hr. The slow growth of mutants generated via allelic exchange confirmed that the ribosomal lesions directly reduce growth rates (Figure 3—figure supplement 1A). Because conditions that limit bacterial growth have been show to reduce the bactericidal activity of antibiotics (Tuomanen et al., 1986), we examined whether the observed shifts in antibiotic sensitivity could be a simple consequence of slowed bacterial growth. To directly test whether slow growth itself was causing the MIC shifts, we manipulated the growth rate of wild-type M. smegmatis by altering the temperature or restricting the carbon source in the growth medium (Figure 3B, Figure 3—figure supplement 1B). Slower growth rates alone did not recapitulate the IC50 shifts seen in the ribosomal mutants. Additionally, we looked to see whether the growth rates in our panel of mutants were correlated with the magnitude of the changes in antibiotic susceptibility (Figure 3—figure supplement 1C). Although growth was correlated with INH resistance (R2?=?0.48) it was actually weakly inversely correlated with CIP and novobiocin (NOV) resistance (R2?=?0.13, 0.14), showing that the higher IC50s of the mutants are cannot be accounted for solely by their slower growth.

Figure 3 with 1 supplement see all

Finally, in order to test whether impaired translation by mutant ribosomes could explain the shifts in antibiotic sensitivity, we inhibited ribosome function with six protein synthesis inhibitors: chloramphenicol, erythromycin, gentamycin, puromycin, streptomycin, and tetracycline. These drugs associate with the ribosome at different locations, interfering with various steps in translation (Yonath, 2005). If resistance were a general consequence of impaired translation, all would display an antagonistic relationship with CIP and INH. We performed 2D checkerboard experiments (Figure 3C), diluting a protein synthesis inhibitor horizontally and a test antibiotic (CIP or INH) vertically across 384-well plates, with the pattern of outgrowth indicating how the two antibiotics interact (DePristo et al., 2007; Yeh et al., 2009). Most protein synthesis inhibitors tested behaved additively with CIP or INH, thus ruling out general impairment of protein synthesis as the likely mechanism of resistance in these mutants. Only tetracycline significantly suppressed the activity of INH (Figure 3D) and CIP (Figure 3E). Among the inhibitors tested, tetracyclines act uniquely by inhibiting tRNA entry at the A-site (Connell et al., 2003), suggesting that this step in protein synthesis may be a common impairment among our ribosomal mutants, and that the consequences of such impairment are transduced to produce multidrug resistance.

### Genome-Scale studies illuminate the effects of ribosomal mutations on ribosome assembly

We used an unbiased approach to reveal changes in the baseline transcriptome and proteome of the ribosomal mutants that might account for the changes in antibiotic and stress sensitivity. We used RNAseq (Pinto et al., 2011) to characterize the baseline log-phase transcriptome of nine ribosomal mutants representing each of the major ribosomal loci, an lfrR mutant., and the parental mc2155 (pUV3583cGFP) strain. Using DESeq2, we identified genes in each mutant that were differentially regulated relative to the parent. All of the ribosomal mutants showed abundant changes relative to the parent; on average, 1023 of 6718 transcripts (15%) were differentially expressed (range 471–2046), and these changes were highly overlapping as demonstrated by a core set of 227 genes that were significantly regulated in all nine ribosomal mutants (Figure 4—figure supplement 2). In contrast, of the 407 genes that were differentially regulated in the lfrR mutant, only 12 were among the core 227 shared by all of the ribosomal mutants. The similarity in behaviour among the different mutants as illustrated by their highly correlated expression profiles (Figure 4A, Figure 4—figure supplement 1) suggests a shared common mechanism for their multi-drug resistance phenotype.

Figure 4 with 3 supplements see all

In parallel, we used a comparative proteomics strategy, iTRAQ (isobaric tags for relative and absolute quantitation) (Ross et al., 2004), to identify proteomic differences between one representative ribosomal mutant, the rplO-1 mutant, and its wild type parent. The overlap between the changes in proteome of the rplO-1 mutant showed moderate correlation with the transcriptomic changes observed by RNAseq (Pearson?=?0.55, p<0.001, Figure 4B). 64 genes were upregulated?>2 fold in both datasets, while 41 were down?>2 fold (Figure 4—figure supplement 3).

This iTRAQ data indicated a likely defect in ribosome assembly as a consequence of the rplO-1 lesion. iTRAQ revealed a reduction in the ratio of large ribosomal subunit proteins relative to small subunit proteins (Figure 4C) that was not reflected in the transcriptomic data, suggesting that the ribosomal mutations might have an effect on the stability or assembly of the large subunit, either decreasing its rate of translation or increasing its rate of degradation. To investigate whether ribosome assembly was perturbed across the various ribosomal mutants, we purified ribosomes from exponentially growing cultures of rplO-1, rplF-1, rpsE-1, rplY-1, and rrlB1 mutants to evaluate the relative abundance of 30S, 50S and 70S subunits using sucrose gradient ultracentrifugation. Indeed, the profiles of the mutants indicate a marked increase in the individual 30S and 50S units and a corresponding decrease in fully assembled 70S subunits relative to the parent (Figure 4D, Figure 4—figure supplement 3).

To examine whether the ribosomal mutation and its impact on the ribosome could be inducing mistranslation as a cause of the multidrug resistance phenotype, we turned to the proteomic data obtained for the rplO-1 mutant compared to its parent. We found 7574 unique peptides with masses corresponding to mistranslated peptides (i.e. the mass corresponds to a known peptide with a single AA change). If mistranslation was occurring in the rplO-1 mutant, these mistranslated peptides should be more relatively abundant in the rplO-1 mutant, which should be reflected in the rplO-1/parent ratio for these peptides. However, the mean and median ratios seen in the mistranslated peptides are no different than the ratios seen across the larger set of ‘normal’ peptide; thus, mistranslated, antibiotic-insensitive targets are unlikely to contribute to resistance. In summary, the ribosomal mutations, while impairing 70S assembly, do not appear to lead to mistranslation; nevertheless, they cause an overall stress to the cell that results in a major reprogramming of the transcriptome and proteome of the bacteria, that ultimately lead to a multi-drug resistant phenotype.

### Genome-Scale studies suggest an overall transcriptional reprogramming contributing to multi-drug resistance

As the ribosomal mutations resulted in significant transcriptional reprogramming of the bacteria, we sought to look for individual clues that might account for the multi-drug resistance phenotype of the mutants. For example, the genome-scale data provided an immediate hypothesis to the nature of the observed INH resistance seen in the ribosomal mutants. Among the genes and gene products underexpressed in all strains by RNAseq and also in the iTRAQ dataset was MSMEG_6384, encoding the catalase KatG, which is required for converting the prodrug INH to its active form Zhang et al. (1992). Relative to the parent strain. levels of katG transcript were reduced 7.1-fold in the rplO-1 mutant (adjusted p-value=1.0×10?54) and in all of the other ribosomal mutants (average of 3.9-fold across all of the mutants, minimum 2.5-fold, adjusted p-value=1.1×10?13), while iTRAQ showed a 4.2-fold reduction at the protein level in the rplO-1 mutant. The changes katG expression were confirmed by qRT-PCR (Figure 4—figure supplement 1) and Western blot analysis (Figure 4E). We hypothesized that among the dramatic overall transcriptome and resulting proteomic changes, the specific reduced KatG levels in the mutants could account for INH resistance. To test this, we introduced a plasmid (pIkatG) carrying an anhydrotetracycline (AHT) inducible copy of katG into an rplO-1 background. We then titrated the concentration of AHT to restore wild-type levels of KatG, measured by Western blot. Indeed, INH sensitivity was restored in the mutant background solely by restoring KatG levels back to wild-type levels, (Figure 4E), suggesting that INH resistance is due to the perturbations in KatG levels caused by the ribosomal mutations on the transcriptome and subsequently, on the proteome.

Similarly, we noted that among the global transcriptomic and proteomic changes occurring in the mutants, the induction of WhiB7, a transcription factor conserved throughout the actinomycetes that positively regulates genes that contribute to innate antibiotic resistance, was among one of the strongest signals. As measured by iTRAQ, WhiB7 was one of the most highly induced proteins (6.0-fold, adjusted p-value=2.5×10?4) and by RNAseq, whiB7 mRNA was up 7.4-fold in rplO-1 relative to the parent strain (p adj.=4.4×10104) and in all of the other ribosomal mutants (average of 3.2-fold; minimum 1.7-fold, adjusted p-value=2.1×10?6). In M. smegmatis, whiB7 has been shown to be highly induced after exposure to specific classes of protein synthesis inhibitors, particularly tetracyclines and macrolides (Burian et al., 2012). We also observed high induction in the mutants of many genes known to be part of the whiB7 regulon, including several transporters that may contribute to antibiotic resistance by export (MSMEG_5187, a known antiporter of tetracycline (De Rossi et al., 1998), and MSMEG_5102). In addition to whiB7, there was strong correlation between the genes induced in our studies and genes previously shown to be upregulated in a strain lacking MSMEG_6129, a putative anti-sigma factor that is an indirect negative regulator of whiB7 (?59) (Figure 4C, highlighted). The changes in whiB7 and katG expression were confirmed by qRT-PCR (Figure 4—figure supplement 1).

In order to investigate the functional contribution of the overexpression of whiB7 to the observed antibiotic resistances, we first deleted the whiB7 gene in both a wild-type and an rplO-1 background. In the rplO-1 background, deletion of whiB7 restored sensitivity to a CIP, ethionamide (ETH) and cycloserine (CYC) (Figure 4C), indicating a critical role for WhiB7 in the broad-spectrum resistance observed in the rplO-1 mutant. Deletion of whiB7 however, did not fully restore INH sensitivity to rplO-1, nor did it alleviate the increased sensitivity to meropenem. We then constitutively expressed WhiB7 at a high-level from a plasmid in a wild-type background and found that it alone did not confer resistance to CIP, INH, ETH, or MER (Figure 4—figure supplement 2). Thus, WhiB7 induction does play a role in the multi-resistance phenotypes; however, it alone is insufficient to account for the entire resistance profiles of the ribosomal mutants. It must act in concert with other pathways to exert its effect on CIP, ETH, MER and INH resistance, with the example of INH resistance resulting via additional alterations in katG expression.

### Ribosomal mutations promote the evolution of high level resistance

We next turned to understand whether ribosomal mutations can contribute to the evolution of high-level multidrug resistance. We first investigated if ribosomal mutants can emerge directly from a multidrug environment, We performed selections in CIP and INH administered in combination (Figure 5A). Mutants arose during double selection at a 3600-fold higher frequency (5.5 × 10?7) than the product of their frequencies in CIP (1.5 × 10?5) and INH (1.0 × 10?5) alone, suggesting that the mutants that arose in the presence of both CIP and INH are not the product of two completely uncoupled mechanisms. When we PCR amplified and sequenced a fragment including the rplF, rpsE and rplO genes from three mutants that arose under double selection, we identified the same frameshift mutations in rplO in two of these that we had previously observed in rplO-2 (77ΔA) and rplO-3 (74-77ΔGTGA).

Figure 5

Ribosomal mutations should also promote the acquisition of high-level antibiotic-specific resistance (e.g., gyrA or lfrR) due to the larger population of surviving cells from which resistant mutants can emerge (Figure 5B). In support of this, we found that upon selection at 3.5 μg/ml CIP, a bactericidal concentration for both wild-type cells and the ribosomal mutants (Figure 5B), strains with mutations in rplO or rplF yielded 10–100 fold more high-level CIP resistant mutants per input cell than wild-type cells (Figure 5C). When we sequenced the gyrA and lfrR loci in these mutants, secondary mutations in both of these genes were identified (Table 2, strains rplO-lfrR-1 through six and rplO-gyrA-1).

Table 2

While the ribosomal mutations result in a fitness cost in antibiotic-free conditions (Figure 3A), acquisition of target- or efflux-associated mutations renders the ribosomal mutations dispensable. Subsequent acquisition of compensatory changes could restore fitness, thereby relegating the ribosomal mutations to a role as an evolutionary ‘bridge’ across periods of exposure to antibiotic or other stress. To test for this possibility, we first sought to determine if compensatory changes that restore fitness could occur in the absence of antibiotic selection. We serially passaged three parallel cultures of the rplO-1 mutant in antibiotic-free medium. By passage 12, cultures demonstrated wild-type growth rates (Figure 5D) as well as wild-type sensitivities to CIP and INH (Figure 5E). Evolved strains in all three cultures had acquired an additional insertion or deletion at the rplO locus (Figure 5F, Table 2), restoring the wild-type reading frame but leaving short deletions or coding changes in the region between the two sequential mutations. We then demonstrated that reversion to wild-type growth rates can also occur readily in a rplO gyrA double mutant (rplO-1 gyrA-1, Table 2) during continued passage at 3.5 μg/ml CIP (Figure 5G–H, Table 2); INH resistance is lost as a consequence of a compensatory one base insertion that restores the original reading frame but alters one codon in rplO. Thus, these secondarily acquired mutations within rplO that restore the reading frame provide a facile, straightforward mechanism for restoring bacterial fitness after a higher level resistance lesion, i.e., gyrA mutation, has been acquired.

Finally, one well in our initial multi-well selection experiments illustrates how rapidly the transition from low to high-level resistance can occur. Fluorescence was detected in well A-G20 after 5 days of CIP exposure, a timing consistent with the slow emergence of ribosomal mutants in this assay. The culture expanded from A-G20 was also resistant to INH, NOV, ETH and CYC, consistent with a ribosomal mutation. However, after further expansion for genomic DNA preparation, WGS revealed SNPs in both lfrR (T452C) and rrlB (C904A). Additionally, MIC assays showed high level resistance to CIP, consistent with the presence of an lfrR efflux mutation. To better understand how these two SNPs came to coexist in the A-G20-derived culture, we colony-purified 23 clones from the frozen stock originally expanded from well A-G20 and used PCR to characterize the rrlB and lfrR loci (Table 3). While the majority (19/23 clones) had only the rrlB C904A mutation, three had both the rrlB C904A and lfrR T452C mutation. Surprisingly, the final clone had the rrlB C904A mutation paired with a different lfrR allele, 63(+C). Thus, the rrlB C904A mutation likely arose first, enabling slow replication in the presence of CIP and accounting for the delayed emergence of well A-G20 on day 5. Continued selective pressure during the remainder of the multiwell assay and during subsequent expansion in 0.3 μg/ml CIP enabled the eventual emergence of lfrR mutants from the rrlB C904A background. These second mutations, providing high level CIP resistance, would increase fitness in the presence of CIP; as seen in Figure 3A, the growth rate of rrlB mutants is slower in the presence of 0.3 μg/ml CIP than in antibiotic-free medium. Thus, a single cycle of selection and expansion was sufficient to demonstrate the evolutionary value of ribosomal mutations to the development of high level resistance.

Table 3

## Discussion

With increasing antibiotic resistance, there is increased urgency to understand how on the molecular and cellular level, bacteria evolve to acquire resistance. Despite strategies such as combination therapy, resistance is nevertheless arising in complex, multidrug environments. In this work we have identified a novel class of mutations in ribosomal subunits and rRNA that surprisingly confers resistance to several mechanistically diverse antibiotics. These alterations in the ribosome do not result in mistranslation, yet cause misassembly of the ribosome, leading to a dramatic shift in the transcriptional program of the bacterium by that imparts resistance to multiple antibiotics with different mechanisms of action. Within the large-scale changes are numerous, discrete changes in individual genes that result in resistance to specific antibiotics. For example, changes in katG expression confer INH resistance while induction of whiB7 plays a role in altering susceptibility to CIP, ethionamide, and cycloserine. Single mutations in the ribosome allow a bacterium, when simultaneously exposed to lethal concentrations of multiple antibiotics (CIP and INH), to survive, grow, and ultimately acquire additional mutations that confer high-level resistance. These mutations are thus beneficial in multidrug environments.

Even the low to moderate levels of resistance in the ribosomal mutants are sufficient to confer a survival advantage over wild-type cells in the presence of antibiotic(s). Whether the mutations allow them to survive in general or simply to prolong their time to death, this extended survival provides the ribosomal mutants an increased opportunity, relative to wild-type bacteria, to acquire a second mutation to become high-level resistant. Thus the presence of a small population of ribosomal mutants could contribute to the survival of the larger population, especially in antibiotic concentrations fluctuating near or above the MIC (as might be seen during treatment of a bacterial infection), and facilitate fixation of the second high-level resistance conferring mutation in the population.

These mutations in the ribosome do carry a fitness cost in the absence of antibiotic. However, the nature of these particular mutations are remarkable because locus-specific reversions can easily occur to restore fitness upon passage in conditions where these mutations no longer confer an advantage. Such conditions include an antibiotic-free environment or the continued presence of an antibiotic but after the acquisition of a second, higher-level resistance-conferring mutation that renders the ribosomal mutation dispensable. The nature of the identified lesions (frameshifts in protein coding genes and substitutions in rrlB) and their respective reversions (restoration of the original reading frame) illustrate the straightforward means for regaining fitness via additional insertions/deletions or via gene conversion. This is particularly facile in the case of the 23S rRNA mutations because like many bacteria, M. smegmatis has multiple rRNA operons (two). Resistance-conferring rrlB mutations were acquired only in one copy of the rRNA, thereby facilitating, RecA-mediated gene conversion from the other, wild-type copy (Prammananan et al., 1999). Similar to the way reversions contribute to the evolution of cefotaxime resistant ?-lactamases by removing mutations that alone increase fitness but are deleterious in combination with certain other fitness-enhancing mutations (DePristo et al., 2007), the ability to revert in this manner greatly expands the number of possible trajectories to resistance. Thus, these particular mutations make accessible an efficient evolutionary path to multidrug resistance via a temporary state of impaired fitness that is ultimately reversed to restore wild-type fitness.

Prokaryotic ribosomes are complex molecular machines in which we now show that alterations can confer resistance to a range of mechanistically diverse antibiotics. The two ribosomal subunits together contain over 50 proteins, the functions of which have not yet been fully elucidated. The majority of these proteins are expected to be essential, although recent studies suggest in E. coli and B. subtilis that as many as 22 may be conditionally dispensable (Shoji et al., 2011; Akanuma et al., 2012). Among these non-essential proteins are two identified in this study that can play a role in multidrug resistance, namely L15 (RplO) and L25 (RplY). The other ribosomal proteins identified in our study, L6 (RplF) and S5 (RpsE), are essential in these organisms. Transposon-based studies of gene essentiality in the related mycobacterium M. tuberculosis (Sassetti et al., 2003; Zhang et al., 2012) show that rplF and rplE are also essential in M. tuberculosis, while rplY was non-essential and the essentiality of rplO was indeterminate. The mutations we found in rplF and rpsE occur near the 3’ end of the open reading frames, leaving the majority of the corresponding proteins, L6 and S5 unaltered.

Our drug interaction experiments using a range of protein synthesis inhibitors potentially link these mutations with the action of tetracycline. Although tetracycline binds to the ribosome at several locations (Day, 1966), its primary binding site is located near the A site, where it interacts with 16S RNA in the 30S subunit (Brodersen et al., 2000). At this site, tetracycline inhibits protein synthesis by interfering with tRNA entry (Wurmbach and Nierhaus, 1983; Maxwell, 1967). Thus, the ribosomal mutations we have identified may have subtle effects in translation initiation that are similar to the effects of low concentrations of tetracycline. These are then sensed by the cell, leading to a large shift in the overall transcriptome, including the specific induction of whiB7 and other specific transcriptional alterations that affect antibiotic sensitivity to other antibiotics. As a corollary, this work raises the question of whether tetracyclines may more generally interact with antibiotics negatively to impact sensitivity.

While traditional culture conditions in the laboratory select for strains with ribosomal properties that maximize growth rates, the varied environments encountered by microbes in nature may select for other ribosomal properties; the slower growth rates of natural isolates of E. coli were shown to be well-correlated with the kinetic properties of their ribosomes, and extended axenic culture of these isolates led to the evolution of culture-adapted strains with growth rates similar to those seen in laboratory strains (Mikkola and Kurland, 1991). The existence of natural bacterial isolates with submaximal ribosome kinetics implies that under certain conditions, suboptimal or altered ribosome function may in fact be beneficial. By linking alterations in ribosome assembly with broad transcriptional reprogramming that enhances survival in fluctuating environments, M. smegmatis and other actinomycetes may benefit from ribosomal variants, even at the cost of decreased maximal growth rate. Via genetically accessible paths to reversion, these populations can serve as stepping-stones to higher levels of drug resistance without any irreversible cost in fitness. Because environmental or pathogenic bacteria face extended or fluctuating periods of stress, the generation of minority populations that are resistant to a broad range of antibiotic and other stresses could serve as an evolutionary bridge to the eventual emergence of fit, antibiotic resistant strains.

## Materials?and?methods

### Strains and media

Request a detailed protocol

Mycobacterium smegmatis mc2155 was grown at 37°C in Middlebrook 7H9 medium (M7H9) supplemented with ADS (Albumin-Dextrose-Saline: Bovine Serum Albumin (Fraction V) 5 g/liter, dextrose 2 g/l, sodium chloride 0.85 g/l final concentration) and Tween 80 (0.05%). LB agar was employed for growth on solid medium. To generate a fluorescent strain, mc2155 was transformed via electroporation with pUV3583cGFP, a derivative of pUV15tetORm (Ehrt et al., 2005) in which the tetracycline regulated promoter Pmyc1 was replaced by a 525 bp fragment of M. tuberculosis gDNA carrying the promoter for the highly-expressed carD gene. Mycobacterium smegmatis mc2155 was obtained from Keith Derbyshire at the University of Albany.

### Selection of resistant mutants

Request a detailed protocol

Bacteria were seeded at a density of 10 organisms/well into 15 μl of M7H9 in clear-bottom plates using a FACS Aria II SORP flow cytometer (Becton Dickinson) and allowed to expand overnight prior to addition of ciprofloxacin at a final concentration of 0.3 μg/ml (1.2X the measured MIC). Plates were sealed with foil and incubated at 37°C without shaking. Growth of resistant mutants was assessed by measuring fluorescence accumulation in each well (bottom read, excitation 485 nm, emission 538 nm, cutoff 530) on a roughly daily basis. Plates were monitored for a total of 12 days. On day 12, fluorescent wells were harvested and expanded to 3 ml in the continued presence of 0.3 μg/ml CIP. An aliquot was frozen at ?80°C and the remaining contents were again expanded to 25 ml in 0.3 μg/ml CIP for the purpose of gDNA preparation.

### Bacterial viability measurements

Request a detailed protocol

The viability of cells during initial selection was measured using a most probable number (MPN) assay (Cochran, 1950). Cells were aliquotted into 4–8 replicate wells in a Costar 96-well black clear bottom plates and then serially diluted ten fold. Plates were grown for approximately 10 days to allow wells in which only single cells were transferred to become turbid. Plates were then read in M5 SpectraMax plate reader (excitation 485, emission 538, cutoff 530) and MPN subsequently calculated. The 95% confidence interval of the MPN assay is a function of the pattern of observed outgrowth. For heat and SDS stress experiments, viability was measures by serial dilution in 4-fold steps in growth medium followed by counting of colonies after spotting 5 μl using a dissecting microscope.

### Antibiotic sensitivities

Request a detailed protocol

To ascertain antibiotic sensitivity 1.41-fold (√2) dilutions of antimicrobials were prepared and seeded with cells at optical densities of 0.02, then grown for 2–3 days. Cell numbers were measured with GFP when available or optical density otherwise. The concentration inhibiting outgrowth to 50% of the no antibiotic control, or the IC50, was determined by fitting a Hill function to the outgrowth curve of the following form:

${y}_{M}=b+\frac{\left(m?b\right)}{1+{\left({x}_{M}/IC50\right)}^{n}}$

where three additional parameters are fit: the maximum outgrowth m, the baseline outgrowth b, and the Hill coefficient n. Fit parameters are constrained such that m > b?>?0, n?>?0, and IC50?>?0. Fitting was performed on log-transformed data, and then transformed back to a linear axis, preventing least-squares fitting from being dominated by expression values near the maximum.

### Sequencing of mutants

Request a detailed protocol

Genomic DNA was extracted using the cetyltrimethylammonium bromide (CTAB)-lysozyme protocol described (van Soolingen et al., 1991). For determination of gyrA and lfrR sequences, these loci were amplified using the primers lfrR F, lfrR R, gyrA QRDR F1 and gyrA QRDR R1and sequenced at the MIT Biopolymers Laboratory. Whole-genome sequencing of an initial set of 8 strains, including A-M17, A-L8, A-A13, A-H2, B-03, B-I24, B-A9 and B-P7, was performed at Texas A and M University using an Illumina GenomeAnalyzer IIx, as described in Ioerger et al. (2010). Fragment libraries were constructed using a genomic DNA sample preparation kit from Illumina, Inc. Sequencing data was collected in paired-end mode (from both ends of each fragment) with a read length of 36 bp. Reads were aligned to the genome of M. smegmatis mc2155 as a reference sequence (Genbank accession number NC_008596.1). Insertions and deletions were determined using a local contig-building algorithm, and then single-nucleotide polymorphisms were identified from a re-alignment of the reads. The depth of coverage ranged between 23x to 42x (mean number of reads covering each site in the genome). Strains A-A15, A-A24, A-G16, A-G20, A-K20, A-K24, A-L8, A-M4, B-O11, B-O18, B-P7, B-P17 were subsequently sequenced at the Broad Institute, also using Illumina materials for library preparation materials and Illumina instrumentation for data generation.

Confirmation of ribosomal lesions by PCR and sequencing was done using the following primers (see Supplementary file 1 for sequences): rplO: rplO F1 and rplO R1; rplF: rplF F1 and rplF R1; rplY: ctc F1 and ctc R1; rpsE: rpsE1 and rpsER1, rplW/B: rplW/B seqF1 and rplW/B seqR1. Because M. smegmatis has two ribosomal RNA operons, rrnA and rrnB, whole genome sequencing data reflected the presence of both wild-type and mutant alleles, but was insufficient to determine which of the two operons contained the mutated rrl. To address this, a conserved primer within the operon, rrl-F1, was paired with either of two unique primers from the region downstream of the operon to amplify the rrlB gene in allele-specific manner (rrlA-R1 or rrlB-R1.) The primers rrl-F2, rrl-F3, rrl-F4, rrl-F5, rrnb_s1, rrnb_s2, rrnb_s3 were used to sequence these products.

### Generation of allelic exchange mutants

Request a detailed protocol

For ribosomal mutants, chromosomal regions from isolate A-A13 (rplO-1) and A-K20 (rpsE-1) were amplified using primers MS_rplO_AX F, MS_rplO_AX R, Ms_rpsE_AX F, and Ms_rpsE_AX R. PCR products were digested with PacI and AscI and cloned into pJG1100, a suicide vector carrying aph (kanamycin resistance), hyg (hygromycin resistance), and sacB (sucrose sensitivity). For whiB7 null mutants, flanking regions were amplified with primers Ms_whiB7KO_USr2_Pac1, Ms_whiB7KO_USf2_Asc1, Ms_whiB7KO_DSf1_Not1, Ms_whiB7KO_DSr1_Pme1 and cloned into pJG1100 in two steps after PacI-AscI and PmeI DraI digestion to yield a vector in which whiB7 is replaced by a hygromycin resistance cassette. After sequence confirmation, plasmids were electroporated into M. smegmatis mc2155, and integrants (single recombinants) were selected on LB containing 50 μg/ml hygromycin B and 12.5 μg/ml kanamycin. Integrants were picked into 7H9 medium and grown overnight prior to plating on LB containing 2.5% sucrose. Colonies were picked into 96 well plates and grown until visibly turbid in 7H9 prior to spotting on LB containing 50 μg/ml hygromycin B (for ribosomal mutants) or kanamycin (for whiB7 null mutant). Hygromycin sensitive colonies were screened for the presence of the mutant allele by PCR amplification of the relevant ribosomal gene. Complementation of the rplO-1 allelic exchange mutant was carried out by transforming that strain with pUV15tet-rplO, a pUV15tetORm derivatave in which rplO expression is controlled via a tetracycline-regulated promoter. The M. smegmatis rplO open reading frame was PCR-amplified using the primers MS_rplO_5'_wt and MS_rplO_orf_3' and this fragment was digested with PacI and EcoRV and cloned into similarly-cut pUV15tetORm. Transformants were selected on hygromycin and production of functional L15 protein was induced using 50 ng/ml anhydrotetracycline (AHT).

### Measurement of replication rates

Request a detailed protocol

We measured the growth rates of our ribosomal mutants in microtiter plates, the environment in which the mutations were generated, using a serial dilution method. Aliquots of each strain were diluted in a two-fold series across a 384-well plate and the accumulation of fluorescence in each well was assessed automatically approximately every hour. We then determined the time at which each dilution crossed a threshold, roughly 3-fold above the auto-fluorescence background. The delay in crossing this threshold between adjacent twofold dilutions reflects the doubling time in this format. To confirm that allelic exchange mutants showed similar impairments in growth to the original mutants, growth rates were assessed by periodic sampling of exponentially growing 30 ml cultures.

### Analysis of ribosome populations

Request a detailed protocol

100 ml of culture at OD600?=?0.4–0.6 was treated with 50 μg/ml of chloramphenicol (CAM) to freeze ribosome activity. Cells were pelleted and resuspended in buffer TM (20 mM Tris pH7.5, 15 mM MgCl2, 1 mg/ml PMSF, 50 μg/ml CAM). Cells were lysed by bead beating. 20 μl of 10% deoxycholate was added to lysates, which were clarified at 15,000 rpm for 15 min. 5 mg of cleared lysate (500 μl of 10 mg/ml cleared lysate) was layered onto a 10–40% sucrose gradient in Buffer E (10 mM Tris, 10 mM MgCl2, 100 mM NH4Cl, 3 mM β-mercaptoethanol, 50 ug/ml CAM). The sucrose gradient was spun in a Beckman ultracentrifuge at 35,000 rpm (150,000xg)?for 2.5 hr at 4°C. The gradient was fractionated using a gradient fractionator (BioComp Instruments, Inc., NB, Canada).

### Checkerboard experiments

Request a detailed protocol

The ability of small-molecule protein synthesis inhibitors to phenocopy the effects of ribosomal mutations was assayed in 384 well microtiter plates, using M. smegmatis (pUV3583cGFP), with fluorescence as a readout. Along the short axis of two plates (31 steps, with no antibiotic in the final row of plate 2) a readout antibiotic (or SDS) was diluted such that each successive wall was 0.8X as concentrated as the next. Likewise, along the long axis (23 steps, with no antibiotic in column 24), a protein synthesis inhibitor was similarly diluted. Readout antibiotics included CIP, INH, MER, and streptomycin (SM), and modulator antibiotics including tetracycline (TET), erythromycin (ERY), SM, puromycin (PUR) and gentamycin (GM).

A common metric used to define drug-drug interactions is Loewe additivity, a concept that stems from the notion that a given drug should neither synergise nor antagonize when placed in combination with itself. While intuitive, competing notions of additivity, such as Bliss independence, do not follow this intuitive idea because the response of bacteria to drugs can be highly nonlinear.

We define the additive curve between two drugs, DrugX and DrugY, as a linear function of the form:

${c}_{DrugY}=m?{c}_{DrugX}+b$

where m is the slope, b is the curve’s intercept with the y-axis, and [cDrugX] and [cDrugY] are concentrations of the two drugs. We constrain this curve by demanding that when DrugX is absent (i.e., [cDrugX]?=?0) the concentration of DrugY is equal to its IC50 concentration (IC50DrugY), and vice versa. Using these as boundary conditions we solve for the intercept, b:

$b=IC5{0}_{DrugY}$

and the slope, m:

$m=?\frac{IC5{0}_{DrugY}}{IC5{0}_{DrugX}}$

so that finally

$\left[{c}_{DrugY}\right]=?\frac{IC5{0}_{DrugY}}{IC5{0}_{DrugX}}?\left[{c}_{DrugX}\right]+IC5{0}_{DrugY}$

In the limiting case where DrugX and DrugY are the same chemical compound (so a checkerboard experiment is performed with the same drug diluted in both axes) then $IC50DrugX≡IC50DrugY$. The equation above then reduces to the simple form:

$\left[{c}_{DrugY}\right]+\left[{c}_{DrugX}\right]=IC5{0}_{DrugY}$

which satisfies Loewe’s criterion where a single drug has no synergistic or antagonistic effect on itself. In our figures the additive curve appears as an arc, not a straight line, because the both axes are displayed in a log scale.

### Transcriptional analyses

Request a detailed protocol

### Proteomic analyses

Request a detailed protocol

### Determination of KatG levels

Request a detailed protocol

Cultures of mc2155, ribosomal mutants, and ribosomal allelic-exchange mutants inducibly expressing MSMEG_6384 were grown in the absence of antibiotic to OD600 0.8 before being diluted to 0.25 and subjected to 6 hr induction with the indicated concentration of anhydrotetracycline. The cultures were collected by centrifugation then washed twice in a full volume of iced PBS?+?0.05% Tween 80, resuspended in 400 μl extraction buffer (50 mM Tris-HCl pH 7.5, 5 mM EDTA, 0.6% SDS, 1 mM PMSF), and bead beaten with 200 μl 0.1 mm Zirconia/Silica beads (BioSpec) for a total of?3 min with 1 min on ice between pulses. Protein extracts were quantified using a standard BCA Protein Assay (Pierce), and 12.8 μg of extract was loaded on 10% Mini-PROTEAN TGX precast gels (Bio-Rad) and run at 200?V; equal loading between samples was assessed on a gel run in parallel using Coomassie blue staining. Electrophoretic transfer of the protein bands to PVDF membrane (Immun-Blot PVDF, Bio-Rad) was accomplished at 100?V for 60 min on ice, followed by blocking in TBST with 5% powdered milk for 60 min at room temperature with agitation. All antibodies used were diluted into TBST containing 1% powdered milk, and all incubation steps were completed on a rocking platform. Membranes were incubated with primary antibody (1:5,000) overnight at 4°, followed by 3 × 10 min washing in TBST, and 3.5 hr incubation at room temperature with secondary antibody (1:10,000). Primary antibody was obtained through the NIH Biodefense and Emerging Infections Research Resources Repository, NIAID, NIH: Monoclonal Anti-Mycobacterium tuberculosis KatG (Gene Rv1908c), Clone IT-57 (CDA4), NR-13793 RRID:AB_2631999. Secondary HRP-labelled goat anti-mouse antibody was acquired from Santa Cruz Biotechnology, Cat# sc-2005 RRID:AB_631736. Blots were developed using Western Lightning Plus-ECL (PerkinElmer). Results were confirmed in a repeated experiment, and data used in the figures is indicative of the initial experiment only.

Inducible expression of KatG was achieved by introduction of a plasmid, piKatG, into the rplO-1 AX1G allelic exchange mutant by electroporation, since the original isolate already carried an incompatible plasmid for GFP expression. piKatG is a derivative of pUV15tetORm in which the gfp ORF is replaced with a PCR product containing the MSMEG_6834 ORF. Primers were MSMEG_6834_Pac1-F and MSMEG_6834_EcoRV-R, and the resulting product was introduced into pUV15tetORm after digestion of each with PacI and EcoRV. pUV15tetORm and its derivatives express TetR, allowing for anhydrotetracycline-inducible expression of genes in Mycobacterium from a strong promoter into which tet operator sites have been engineered.

## References

1. 1
2. 2
3. 3
4. 4
Antibiotic-selective environments (1998)
Clinical Infectious Diseases 27 Suppl 1:S5–S11.
https://doi.org/10.1086/514916
5. 5
Strategies to minimize the development of antibiotic resistance
(1997)
Journal of Chemotherapy 9 Suppl 3:29–37.
6. 6
7. 7
8. 8
9. 9
10. 10
11. 11
12. 12
Antibiotic Resistance Threats in the United States, 2013
(2013)
Antibiotic Resistance Threats in the United States, 2013.
13. 13
14. 14
15. 15
16. 16
17. 17
18. 18
Vancomycin resistance in gram-positive cocci (2006)
Clinical Infectious Diseases 42 Suppl 1:S25–S34.
https://doi.org/10.1086/491711
19. 19
Tetracycline inhibition of cell-free protein synthesis. I. binding of tetracycline to components of the system
(1966)
Journal of Bacteriology 91:1917–1923.
20. 20
21. 21
Molecular cloning and functional analysis of a novel tetracycline resistance determinant, tet(V), from Mycobacterium smegmatis
(1998)
Antimicrobial Agents and Chemotherapy 42:1931–1937.
22. 22
23. 23
The mutant selection window and antimicrobial resistance (2003)
Journal of Antimicrobial Chemotherapy 52:11–17.
https://doi.org/10.1093/jac/dkg269
24. 24
25. 25
26. 26
27. 27
28. 28
29. 29
30. 30
31. 31
32. 32
Mechanisms of resistance to quinolones (2005)
Clinical Infectious Diseases 41 Suppl 2:S120–S126.
https://doi.org/10.1086/428052
33. 33
34. 34
35. 35
36. 36
The population dynamics of antimicrobial chemotherapy
(1997)
Antimicrobial Agents and Chemotherapy 41:363–373.
37. 37
38. 38
39. 39
Mutations of bacteria from virus sensitivity to virus resistance
(1943)
Genetics 28:491–11.
40. 40
41. 41
42. 42
43. 43
44. 44
How drug resistance emerges as a result of poor compliance during short course chemotherapy for tuberculosis
(1998)
The International Journal of Tuberculosis and Lung Disease 2:10–15.
45. 45
46. 46
47. 47
Antimicrobial agent resistance in mycobacteria: molecular genetic insights
(1995)
Clinical Microbiology Reviews 8:496–514.
48. 48
49. 49
Tackling drug-resistant infections globally: An overview of our work
(2016)
London, UK: Resistance RoA.
50. 50
51. 51
52. 52
53. 53
Efflux-mediated antimicrobial resistance (2005)
Journal of Antimicrobial Chemotherapy 56:20–51.
https://doi.org/10.1093/jac/dki171
54. 54
RecA-mediated gene conversion and aminoglycoside resistance in strains heterozygous for rRNA
(1999)
Antimicrobial Agents and Chemotherapy 43:447–453.
55. 55
56. 56
57. 57
58. 58
59. 59
60. 60
61. 61
62. 62
63. 63
64. 64
65. 65
66. 66
67. 67
68. 68
69. 69
Occurrence and stability of insertion sequences in Mycobacterium tuberculosis complex strains: evaluation of an insertion sequence-dependent DNA polymorphism as a tool in the epidemiology of tuberculosis
(1991)
Journal of Clinical Microbiology 29:2578–2586.
70. 70
71. 71
72. 72
73. 73
74. 74
75. 75
76. 76
77. 77

## Decision letter

1. Michael S Gilmore
Reviewing Editor; Harvard Medical School, United States

In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.

Thank you for submitting your article "Ribosomal mutations promote the evolution of antibiotic resistance in a multidrug environment" for consideration by eLife. Your article has been favorably evaluated by Richard Losick (Senior Editor) and three reviewers, one of whom, Michael S. Gilmore (Reviewer #1), is a member of our Board of Reviewing Editors. The following individuals involved in review of your submission have also agreed to reveal their identity: Helena Boshoff (Reviewer #2) and William R. Jacobs Jr. (Reviewer #3).

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

The manuscript "Ribosomal mutations promote the evolution of antibiotic resistance in a multidrug environment" by Gomez et al. investigates the mechanism by which mutants of Mycobacterium smegmatis, as a model for Mycobacterium Tuberculosis, acquire resistance to the antibiotic ciprofloxacin upon long-term exposure at 1.2 X MIC. It is state-of-the-art, interesting, and provocative. The mutant selection is innovative, as the authors set up mini-cultures in 384 well plates seeded with 10 cells in each well. When the population in each well reaches 105 cells per ml, ciprofloxacin is added. This design ensures that the mutants analyzed from different wells are not siblings. Of those, 19 mutants were screened for polymorphisms in gyrA, and interestingly no mutations mapped to this gene. Whole genome sequencing was then performed on the 19 independent mutants. Four mutations map to the repressor of the LfrA efflux pump. Of the 15 remaining mutants, 14 possessed mutations in ribosomal genes. The authors generate specific mutations in the parental strain, and indeed confirm the mutations mediate the phenotypes. Interestingly, these mutations confer resistance to numerous other antibiotics that have notably different mechanisms of action, with the greatest cross-resistances to isoniazid, detergent and heat stress. Not only are the results interesting, the approach is innovative taking advantage of modern high throughput methodologies and will become a blueprint for future studies.

In review of the manuscript, two major concerns were noted: For technical reasons, a potential misinterpretation of the data cannot be excluded. The parental strain is actually not wild type, but rather a mutant of M. smegmatis ATCC 607 that was first isolated in 1988 (Snapper et al. PNAS 85: 6987-6991). The efficient plasmid transformation phenotype of this strain relates to its being permissive for plasmid replication (Snapper et al. 1990 Mol. Microbiol. 11:1911-1919). Recently, it was demonstrated that this property maps to a gene encoding an SMC protein (Panas et al. 2014 PNAS 111:13264-13271). The precise mechanism by which SMC allows plasmid replication has yet to be elucidated, but in wild type strains it either prevents replication or plasmid segregation to daughter cells. The authors use a concentration of ciplofloxocin that may not be bactericidal (in the second paragraph of the subsection “Selection for Low Level Ciprofloxacin Resistance Identifies Ribosomal Mutants”, it is stated that ciprofloxacin has a relatively slow killing rate – the data should be included as a supplemental figure), and the surrogate measure of escape from killing is indirect and relies on plasmid replication to measure green florescent protein activity. Therefore, it is possible, perhaps likely, that ciprofloxicin at the level being used may be limiting plasmid replication in the cell rather than selecting for resistance to killing, and the mutations obtained relieve that inhibition. It is therefore important to distinguish between allowing the pAL5000 replicon to replicate and cell death (It is ambiguous whether the GFP tag is on a self-replicating plasmid or integrated. Loss of the plasmid from the cell would be scored as a dead cell in most of the assays. Even if an integrated plasmid was used, excision could be dependent on the topology of the DNA altered by fluoroquinolones. In the Methods, it is important to note whether the selecting antibiotic is included during the experiments.) Several approaches might be useful in resolving the ambiguity, including directly assessing killing; using the parent of mc2155, mc26, to see if the mutations identified confer resistance to bactericidal killing, such as engineering the rplO and rplJ mutations in mc26, etc.

A second major concern is that a mechanism that confers resistance to both ciprofloxacin and isoniazid or ethionomide simultaneously is difficult to imagine. The authors mention that the proteomics, conducted on a single ribosomal mutant, rplO, did not indicate increased mistranslation as being a driving factor for these differences. However, it is not clear that the technique would be sensitive enough to detect mistranslation differences that could still drive increases in drug resistance, as reported by for example Su et al. (2016) and Leng et al. (2015). The transcriptional profile for rplF and rplO mutants are almost identical, which is perhaps not totally surprising since both of these are 23S rRNA interacting in the aminoacyltransferase site. (Why rplO and rplF mutants were chosen for the focus of further investigation is not entirely clear. Are there data to suggest that the other mutations in the large subunit or small subunit would, or would not phenocopy these mutants? In the one example of evolved high-level resistance in a single well (Figure 5), the mutation evolves following a mutation to a ribosome-associated gene (rrlB), which is not one of the two classes selected for further characterization.) Epistasis experiments demonstrate that all of these ribosomal mutants show increased WhiB7 expression (a transcription factor that may increase antibiotic efflux), and to some extent decreased KatG expression – two separate mechanisms by which resistance to different classes of antibiotic may occur. WhiB7 has been explored quite a bit in the literature due to its role in conferring multi-drug resistance and its upregulation by various drugs, especially tetracyclines, has previously been reported (Burian et al., 2012). What is missing is determination of the precise mechanism by which these mutations cause the specific upregulation of WhiB7 expression. Knowing the mechanism would represent a major advance.

https://doi.org/10.7554/eLife.20420.024

## Author response

[…] In review of the manuscript, two major concerns were noted: For technical reasons, a potential misinterpretation of the data cannot be excluded. The parental strain is actually not wild type, but rather a mutant of M. smegmatis ATCC 607 that was first isolated in 1988 (Snapper et al. PNAS 85: 6987-6991). The efficient plasmid transformation phenotype of this strain relates to its being permissive for plasmid replication (Snapper et al. 1990 Mol. Microbiol. 11:1911-1919). Recently, it was demonstrated that this property maps to a gene encoding an SMC protein (Panas et al. 2014 PNAS 111:13264-13271). The precise mechanism by which SMC allows plasmid replication has yet to be elucidated, but in wild type strains it either prevents replication or plasmid segregation to daughter cells. The authors use a concentration of ciplofloxocin that may not be bactericidal (in the second paragraph of the subsection “Selection for Low Level Ciprofloxacin Resistance Identifies Ribosomal Mutants”, it is stated that ciprofloxacin has a relatively slow killing rate – the data should be included as a supplemental figure), and the surrogate measure of escape from killing is indirect and relies on plasmid replication to measure green florescent protein activity. Therefore, it is possible, perhaps likely, that ciprofloxicin at the level being used may be limiting plasmid replication in the cell rather than selecting for resistance to killing, and the mutations obtained relieve that inhibition. It is therefore important to distinguish between allowing the pAL5000 replicon to replicate and cell death (It is ambiguous whether the GFP tag is on a self-replicating plasmid or integrated. Loss of the plasmid from the cell would be scored as a dead cell in most of the assays. Even if an integrated plasmid was used, excision could be dependent on the topology of the DNA altered by fluoroquinolones. In the Methods, it is important to note whether the selecting antibiotic is included during the experiments.) Several approaches might be useful in resolving the ambiguity, including directly assessing killing; using the parent of mc2155, mc26, to see if the mutations identified confer resistance to bactericidal killing, such as engineering the rplO and rplJ mutations in mc26, etc.

Major point 1: potential issues surrounding the use of mc2155.

The first major concern related to the genetic background of the strain used for all of our studies, M. smegmatis mc2155. We now include data showing that indeed, the phenomenon we describe for mc2155 also applies to the parental mc26 strain and is thus not an artifact of the genetics of mc2155. Specifically, we have addressed this issue on several levels:

1) While the emergence of the resistant mutants were detected based on GFP fluorescence, microtiter plate-based MICs in several instances were measured by optical density, which mitigates concerns that the “resistance” observed in these mutants is due simply to altered (GFP) plasmid retention.

2) Nonetheless, we have now obtained mc26 (parental strain) and used this strain in a de novo selection experiment to show that CIP-resistant ribosomal mutants could arise spontaneously in this true WT background in the same manner as in mc2155.

3) Finally, we have now repeated the microtiter-plate based selection process with the original mc2155 strain carrying the episomal GFP plasmid and with mc26 and found that there is no significant plasmid loss. Both of these selections were conducted in the absence of hygromycin and survival enumerated in the presence and absence of hygromycin (described in detail in Figure 1—figure supplement 2) thereby enabling us to directly assess the bactericidal effect of ciprofloxacin and the effects of ciprofloxacin stress on plasmid retention in the mc2155 strain. There is no significant plasmid loss. Most importantly, we identified two rplO mutants that arose from the mc26 background during CIP selection, and both of these were also cross-resistant to INH showing that the same phenomenon that we have described for mc2155 holds true for mc26.

A second major concern is that a mechanism that confers resistance to both ciprofloxacin and isoniazid or ethionomide simultaneously is difficult to imagine. The authors mention that the proteomics, conducted on a single ribosomal mutant, rplO, did not indicate increased mistranslation as being a driving factor for these differences. However, it is not clear that the technique would be sensitive enough to detect mistranslation differences that could still drive increases in drug resistance, as reported by for example Su et al. (2016) and Leng et al. (2015). The transcriptional profile for rplF and rplO mutants are almost identical, which is perhaps not totally surprising since both of these are 23S rRNA interacting in the aminoacyltransferase site. (Why rplO and rplF mutants were chosen for the focus of further investigation is not entirely clear. Are there data to suggest that the other mutations in the large subunit or small subunit would, or would not phenocopy these mutants? In the one example of evolved high-level resistance in a single well (Figure 5), the mutation evolves following a mutation to a ribosome-associated gene (rrlB), which is not one of the two classes selected for further characterization.) Epistasis experiments demonstrate that all of these ribosomal mutants show increased WhiB7 expression (a transcription factor that may increase antibiotic efflux), and to some extent decreased KatG expression – two separate mechanisms by which resistance to different classes of antibiotic may occur. WhiB7 has been explored quite a bit in the literature due to its role in conferring multi-drug resistance and its upregulation by various drugs, especially tetracyclines, has previously been reported (Burian et al., 2012). What is missing is determination of the precise mechanism by which these mutations cause the specific upregulation of WhiB7 expression. Knowing the mechanism would represent a major advance.

Major point 2: issues surrounding a common resistance mechanism to multiple drugs (ciprofloxacin and isoniazid).

Indeed, the major finding of this manuscript that we wish to report is both surprising and remarkable, that a single mutation can result in resistance to antibiotics with different mechanisms of action. We apologize if our discussion of the data providing insight into the biology and mechanism underlying this phenomenon was not clear in the original manuscript, thereby preventing the reviewers from understanding what we believe is going on. We have tried to clarify this in the current draft.

1) Based on transcriptomic and proteomic analysis of the mutants, we conclude that the perturbations in ribosomal function due to these mutations leads to a large-scale reorganization of the transcriptome and the proteome. Changes in specific genes within the large-scale reorganization can be used to account for the individual resistance phenotypes that we see to individual antibiotics. For example, a repression of KatG expression accounts for the observed resistance to isoniazid while the upregulation of WhiB7 contributes to ciprofloxacin resistance. We have tried to clarify that changes in specific genes within the global shifts can account for resistance to different antibiotics (subsection “Genome-Scale Studies suggest an overall transcriptional reprogramming contributing to multi-drug resistance”, first two paragraphs). In addition, we now provide additional data to suggest that the global changes that occur in the ribosomal mutants may be occur as a result of the ribosomal mutations causing stress to the bacterial cell due to defective assembly of the large 70S ribosome, thereby providing further molecular insight into the consequences of carrying these described ribosomal mutations (Figure 4C-D, subsection “Genome-Scale Studies Illuminate the effects of ribosomal mutations on ribosome assembly”, first three paragraphs).

2) The question of the mutations causing mistranslation as a cause of resistance was raised. Several lines of evidence argue against mistranslation as the underlying cause of increased resistance to antimicrobials.

In M. smegmatis, mistranslation has been clearly linked to rifampin resistance; we, however, see no shift in sensitivity to rifampin.

Deletion of whiB7 eliminates the resistance to several of the antibiotics, which should not occur if mistranslated targets are the source of resistance.

The question was raised whether the proteomic data was sensitive enough to detect mistranslated peptides. We now have clarified in the last paragraph of the subsection “Genome-Scale Studies Illuminate the effects of ribosomal mutations on ribosome assembly”, that the method is indeed sensitive enough as we are able to detect a significant number of mistranslated peptides even in wild-type. Importantly, there are no differences in numbers of mistranslated peptides between the ribosomal mutant and its parent. Therefore, mistranslation cannot account for the multidrug phenotype.

3) The concern was raised whether we could truly infer that a common phenomenon was occurring across all of the ribosomal mutants that was leading to the shared resistance phenotypes as we had only characterized one rplO and one rplF mutant. To address this, we have expanded our unbiased transcriptomic analysis to include mutants in all loci identified (9 mutants in total), including multiple rplO, rpsE, and 23S rRNA mutants, along with mc2155 and an lfrR mutant for comparison. This new, rich dataset shows that all of the mutants are behaving very similarly based on a common transcriptional program and now allows us to identify transcriptional perturbations that are common across all ribosomal mutants sharing these resistance phenotypes, that explain some of the individual resistance mechanisms (i.e., downregulation of katG to explain isoniazid resistance and WhiB7 induction to explain ciprofloxacin resistance). This new data strongly supports the notion that a common, highly altered transcriptional program relative to wild-type underlies the resistances observed in this manuscript.

https://doi.org/10.7554/eLife.20420.025

## Article and author information

### Author details

1. #### James E Gomez

The Broad Institute of MIT and Harvard, Cambridge, United States
##### Contribution
JEG, Conceptualization, Data curation, Formal analysis, Investigation, Visualization, Methodology, Writing—original draft, Writing—review and editing
##### Contributed equally with
Benjamin B Kaufmann-Malaga
##### Competing interests
The authors declare that no competing interests exist.
2. #### Benjamin B Kaufmann-Malaga

1. The Broad Institute of MIT and Harvard, Cambridge, United States
2. Department of Molecular Biology and Center for Computational and Integrative Biology, Massachusetts General Hospital, Boston, United States
3. Department of Microbiology and Immunobiology, Harvard Medical School, Boston, United States
##### Contribution
BBK-M, Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Writing—original draft
##### Contributed equally with
James E Gomez
##### Competing interests
The authors declare that no competing interests exist.
3. #### Carl N Wivagg

1. The Broad Institute of MIT and Harvard, Cambridge, United States
2. Department of Molecular Biology and Center for Computational and Integrative Biology, Massachusetts General Hospital, Boston, United States
##### Contribution
CNW, Conceptualization, Investigation, Methodology
##### Competing interests
The authors declare that no competing interests exist.
4. #### Peter B Kim

The Broad Institute of MIT and Harvard, Cambridge, United States
##### Contribution
PBK, Conceptualization, Investigation, Methodology
##### Competing interests
The authors declare that no competing interests exist.
5. #### Melanie R Silvis

The Broad Institute of MIT and Harvard, Cambridge, United States
##### Contribution
MRS, Investigation
##### Competing interests
The authors declare that no competing interests exist.
6. #### Nikolai Renedo

The Broad Institute of MIT and Harvard, Cambridge, United States
##### Contribution
NR, Investigation
##### Competing interests
The authors declare that no competing interests exist.
7. #### Thomas R Ioerger

Department of Computer Science, Texas A&M University, College Station, United States
##### Contribution
TRI, Data curation, Formal analysis, Investigation
##### Competing interests
The authors declare that no competing interests exist.

The Broad Institute of MIT and Harvard, Cambridge, United States
##### Contribution
RA, Data curation, Formal analysis, Investigation, Methodology
##### Competing interests
The authors declare that no competing interests exist.
9. #### Jonathan Livny

The Broad Institute of MIT and Harvard, Cambridge, United States
##### Contribution
JL, Data curation, Formal analysis, Investigation, Methodology
##### Competing interests
The authors declare that no competing interests exist.
10. #### Skye Fishbein

Department of Immunology and Infectious Diseases, Harvard T.H. Chan School of Public Health, Boston, United States
##### Contribution
SF, Formal analysis, Investigation
##### Competing interests
The authors declare that no competing interests exist.
11. #### James C Sacchettini

Department of Biochemistry and Biophysics, Texas A&M University, College Station, United States
##### Contribution
JCS, Resources, Formal analysis
##### Competing interests
The authors declare that no competing interests exist.
12. #### Steven A Carr

The Broad Institute of MIT and Harvard, Cambridge, United States
##### Contribution
SAC, Resources, Methodology
##### Competing interests
The authors declare that no competing interests exist.
13. #### Deborah T Hung

1. The Broad Institute of MIT and Harvard, Cambridge, United States
2. Department of Molecular Biology and Center for Computational and Integrative Biology, Massachusetts General Hospital, Boston, United States
3. Department of Genetics, Harvard Medical School, Boston, United States
##### Contribution
DTH, Conceptualization, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Writing—review and editing
##### For correspondence
hung@molbio.mgh.harvard.edu
##### Competing interests
The authors declare that no competing interests exist.

### Funding

#### Doris Duke Charitable Foundation (2008046)

• Deborah T Hung

#### Pershing Square Foundation

• Deborah T Hung

#### Broad Institute Tuberculosis Donor Group

• Deborah T Hung

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

### Acknowledgements

We thank W Thompson-Butler and Melissa Choi for technical assistance, S A Stanley, S Grant, and A Barczak for helpful discussions. We thank J Stokes for extremely helpful input on ribosome assembly and Keith Derbyshire for providing M. smegmatis mc26. BBK thanks the Massachusetts General Hospital ECOR and NYCT Heiser Postdoctoral Fellowships for support. This work was funded by a Doris Duke Clinical Scientist Developmental award (DTH), the Broad Institute Tuberculosis donor group, and the Pershing Square Foundation.

### Reviewing Editor

1. Michael S Gilmore, Harvard Medical School, United States

### Publication history

2. Accepted: January 20, 2017
3. Version of Record published: February 21, 2017 (version 1)

? 2017, Gomez et al.

## Metrics

• 5,090
Page views
• 975
• 20
Citations

Article citation count generated by polling the highest count across the following sources: Scopus, Crossref, PubMed Central.

A two-part list of links to download the article, or parts of the article, in various formats.

### Open citations (links to open the citations from this article in various online reference manager services)

1. Evolutionary Biology
2. Microbiology and Infectious Disease

# A unique chromatin profile defines adaptive genomic regions in a fungal plant pathogen

David E Cook et al.
Research Article

Genomes store information at scales beyond the linear nucleotide sequence, which impacts genome function at the level of an individual, while influences on populations and long-term genome function remains unclear. Here, we addressed how physical and chemical DNA characteristics influence genome evolution in the plant pathogenic fungus Verticillium dahliae. We identified incomplete DNA methylation of repetitive elements, associated with specific genomic compartments originally defined as Lineage-Specific (LS) regions that contain genes involved in host adaptation. Further chromatin characterization revealed associations with features such as H3 Lys-27 methylated histones (H3K27me3) and accessible DNA. Machine learning trained on chromatin data identified twice as much LS DNA as previously recognized, which was validated through orthogonal analysis, and we propose to refer to this DNA as adaptive genomic regions. Our results provide evidence that specific chromatin profiles define adaptive genomic regions, and highlight how different epigenetic factors contribute to the organization of these regions.

1. Evolutionary Biology
2. Genetics and Genomics

# A putative origin of the insect chemosensory receptor superfamily in the last common eukaryotic ancestor

Richard Benton et al.
Short Report Updated

The insect chemosensory repertoires of Odorant Receptors (ORs) and Gustatory Receptors (GRs) together represent one of the largest families of ligand-gated ion channels. Previous analyses have identified homologous ‘Gustatory Receptor-Like’ (GRL) proteins across Animalia, but the evolutionary origin of this novel class of ion channels is unknown. We describe a survey of unicellular eukaryotic genomes for GRLs, identifying several candidates in fungi, protists and algae that contain many structural features characteristic of animal GRLs. The existence of these proteins in unicellular eukaryotes, together with ab initio protein structure predictions, provide evidence for homology between GRLs and a family of uncharacterized plant proteins containing the DUF3537 domain. Together, our analyses suggest an origin of this protein superfamily in the last common eukaryotic ancestor.