Transcriptional Attenuation Controls Macrolide Inducible Efflux and Resistance in Streptococcus pneumoniae and in Other Gram-Positive Bacteria Containing mef/mel(msr(D)) Elements

Macrolide resistance, emerging in Streptococcus pneumoniae and other Gram-positive bacteria, is increasingly due to efflux pumps encoded by mef/mel(msr) operons found on discrete mobile genetic elements. The regulation of mef/mel(msr) in these elements is not well understood. We identified the mef(E)/mel transcriptional start, localized the mef(E)/mel promoter, and demonstrated attenuation of transcription as a mechanism of regulation of macrolide-inducible mef-mediated macrolide resistance in S. pneumoniae. The mef(E)/mel transcriptional start site was a guanine 327 bp upstream of mef(E). Consensus pneumococcal promoter -10 (5′-TATACT-3′) and -35 (5′-TTGAAC-3′) boxes separated by 17 bp were identified 7 bp upstream of the start site. Analysis of the predicted secondary structure of the 327 5’ region identified four pairs of inverted repeats R1-R8 predicted to fold into stem-loops, a small leader peptide [MTASMRLR, (Mef(E)L)] required for macrolide induction and a Rho-independent transcription terminator. RNA-seq analyses provided confirmation of transcriptional attenuation. In addition, expression of mef(E)L was also influenced by mef(E)L-dependent mRNA stability. The regulatory region 5’ of mef(E) was highly conserved in other mef/mel(msr)-containing elements including Tn1207.1 and the 5612IQ complex in pneumococci and Tn1207.3 in Group A streptococci, indicating a regulatory mechanism common to a wide variety of Gram-positive bacteria containing mef/mel(msr) elements.


Introduction
Macrolides are characterized by 14-to 16-membered lactone rings with attached amino sugars which confer an affinity for bacterial ribosomes. Binding of the 23S ribosomal subunit by a macrolide molecule effectively blocks de novo synthesis of proteins to inhibit bacterial proliferation. Macrolides are used to treat a wide range of bacterial infections and are particularly important for treatment of infections cause by Gram-positive bacteria. These include, among others, soft tissue infections often caused by Staphylococcus aureus and Streptococcus pyogenes, and invasive diseases caused by Streptococcus pneumoniae. Macrolides are often the drug of choice for empirical treatment of upper respiratory tract infections which are most often caused by S. pneumoniae. Unfortunately, resistance to macrolides in Gram-positive bacteria, especially in streptococci, complicates the treatment of these diseases. The two major mechanisms of macrolide resistance in these organisms are target modification and active efflux. RNA methylases, such as, Erm(A), Erm(B) and Erm(C) modify specific nucleotides in the 23S rRNA and block macrolide binding. Erm-type methylases confer the MLS B phenotype: highlevel resistance to macrolides, lincosamides and streptogramin B. The M phenotype, conferred by macrolide efflux, is generally characterized by lower levels of resistance to 14-and 15-membered macrolides and sensitivity to 16-membered macrolides, lincosamides and streptogramin B.
Macrolide efflux was first described in Staphylococcus epidermidis and was attributed to msr (A), an ATP-binding subunit ABC-transporter lacking a membrane-binding domain or an obvious membrane bound partner [1]. Subsequently, the mef (macrolide efflux) family of genes encoding major facilitator superfamily-type (MFS) efflux pumps were identified in S. pyogenes [2] and S. pneumoniae [3]. Interestingly, mef genes are invariably found upstream and are cotranscribed with a homologue of msr(A), named msr(D) or mel, suggesting a coordinated function between the MFS and the ABC-transporter as a dual efflux pump [4]. Homologues of mef and msr(D) have been identified in a wide range of Gram-positive bacteria. In S. pneumoniae efflux is encoded by one of four mef alleles: mef(A), mef(E), mef(O) or mef(I). The most common in pneumococci is mef(E) which is associated with the msr(D) homolog mel and is located on the 5.5 kb mobile element Mega (Macrolide efflux genetic assembly) [3]. Mef(A) is more commonly associated with group A streptococci (GAS) and is usually carried on phage-related elements. In the pneumococcus, mef(A) is carried on the degenerative phage element Tn1207.1 [5]. mef(I) was reported in pneumococcal isolates from Spain and is located on the 5612IQ complex, so named because of the presence of the chloramphenicol resistance-encoding catQ gene [6].
Macrolide resistance mediated by rRNA methylation or by efflux is inducible by 14-and 15-member ringed macrolides, such as erythromycin and azithromycin, respectively (reviewed in [7]). It is well established that inducible expression of erm genes is controlled by attenuation, usually at the level of translation. Transcripts of inducible erm genes are characterized by the presence of one or two small (8-20 amino acids) leader peptides encoded 5' of the erm structural gene. Macrolide-bound ribosomes pause at a programmed stall site on the leader peptide disrupting the attenuator and allowing formation of the anti-attenuator. We have previously demonstrated that, unlike most erm genes, mef(E) and mel are induced at the level of transcription of the genes as an operon (mef(E)/mel), however the mechanism of induction has not been elucidated [8]. We have also shown that mef(E)/mel is induced weakly by the human and mouse antimicrobial peptides LL-37 and CRAMP-38, respectively [8,9]. This has not been shown for erm genes. The differences in expression of mef(E)/mel and erm genes suggest dissimilar mechanisms control these macrolide resistance genes. The goal of this study was to characterize the mechanism controlling macrolide induction of transcription of mef(E)/mel by identifying its promoter and searching for cis-acting regulatory sequences in the mef(E)/mel transcript. In addition, phylogenetically analyses of mef(E)/mel and the macrolide efflux systems in other pneumococcal and Gram-positive bacteria.

Ethics statement
All isolates used in the study were received from previously existing pneumococcal strain collections. Ethics approval was obtained from the Institutional Review Board (IRB) of the institutions who collected the biological specimens from the human subjects: Emory University, Centers for Disease Control and Prevention or Georgia Department of Human Resources. Ethics approval was not required for this study because all data were anonymized.

Bacterial strains
Streptococcus pneumoniae GA17457 is an erythromycin-resistant, Mega-containing, serotype 19A strain originally isolated from the blood of a patient in the Atlanta metropolitan area in 2001 [3,4]. GA17457 was selected from 147 isolates, including 115 macrolide resistant isolates, most of which were identified through population-based surveillance or carriage studies conducted by the Georgia Emerging Infections Program (GEIP) (S2 Table) [30]. England 14 -9 is an erythromycin resistant, mef(A)-containing serotype 14 strain isolated from a patient in the United Kingdom in 1993 [31] (S2 Table). In addition ten strains were isolated from the nasopharynx of healthy children in the Atlanta metropolitan area [32]. Draft genome sequences were deposited into the National Center for Biotechnology (NCBI) Whole Genome Shotgun database (WGS). Isolate metadata and the NCBI accession number for each genome is provided in S2 Table. Construction of the mef(E)-lacZ reporter strain XZ7042 from GA17457 has been previously described [8,9]. The control strain XZ7049 was generated by insertion of the promoterless lacZ of pPP2 [33] into bgaA of GA17457. The Genbank WGS master accession number for GA17457 draft genome sequence is AILS00000000 (S2 Table).

Construction of mutations in the regulatory region of the mef(E)-lacZ reporter
To generate mutations in the mef(E)/mel 5' regulatory region governing mef(E)-lacZ expression, plasmid pXZ7032 [9] was used as a template for inverse PCR to introduce point mutations and deletions. For example, the putative mef(E)/mel promoter was deleted by amplification of pXZ7032 with overlapping primers 471-777F and 741-777R (S1 Table). The product was DpnI digested and self-ligated to generate plasmid pXB05 which was transformed into GA17457 and selected on 3 μg ml -1 tetracycline to generate the XZ7042 mef(E)-deletion derivative XB05 (Table 1). the kanamycin-resistance marker immediately upstream of wild type Mega. This inserted the kanamycin resistance marker 780 bp upstream of the mef(E)/mel transcriptional start site. Because of the distance between the desired mutation location and the selective marker, recombination events resulting in kanamycin resistance, but without the desired mutation were possible. Therefore, the nucleotide sequence of the entire 1.1 kb region upstream of mef(E) of the resulting mutants were determined to confirm the mutations and that no other changes in the regulatory region occurred. Strain XB29 was selected as the positive control strain for erythromycin susceptibility assays.
To generate mef(E) regulatory mutants, plasmid pXB101 was used as template for inverse PCR using primers designed to create deletions or introduce point mutations (S1 Table). Inverse PCR products were DpnI digested, ligated and transformed into GA17457 as above. Selection for kanamycin resistance, screening for promoter deletion by PCR and verification of the deletion produced the promoter deletion mutant XB30. XB30 was complemented by restoration of the promoter sequence by transformation of XB30 with a PCR product containing the wild type promoter and selected on 1.0 μg ml -1 erythromycin. The sequence of the entire 1.1 kb mef(E) regulatory region and the putative promoter deletion were verified by Sanger DNA sequencing.

Susceptibility assays
The minimum inhibitory concentration (MIC) of erythromycin was determined by Etest as per manufacturer recommendations (AB bioMerieux, Solna, Sweden). To determine the effects of erythromycin induction on resistance, strains were cultured overnight at 37°C in 5% CO 2 on TSA blood agar plates with or without 10 -1 MIC concentration of erythromycin representing inducing and non-inducing conditions, respectively. Cell suspensions from the overnight plates were swabbed onto cation-adjusted Mueller-Hinton agar plus 5% horse blood. Plates were The 5' end of the mef(E)/mel transcript from GA17457 was mapped by 5 0 Rapid Amplification of cDNA ends (5 0 RACE) as has been previously described [35]. Briefly, total RNA was extracted from GA17457 grown in 10 -1 MIC erythromycin (12 μg ml -1 ) using the Qiagen RNAeasy Mini Kit (Qiagen Sciences, Germantown, MD). RNA was reverse transcribed into cDNA using Avian Myeloblastosis Virus (AMV) reverse transcriptase (New England Biolabs, Ipswich, MA) and target-specific primers SC10, SC143 and SC199 designed to produce overlapping fragments covering the entire 1.1 kb between the start of the Mega element and the mef (E) start codon (S1 Table). Reverse transcription reactions were incubated at 55°C for 1 hr and terminated at 65°C for 10 min. Poly-A tails were added to the 5 0 end of cDNAs by incubation with dATP and terminal transferase at 37°C for 20 min. Tailing reactions were terminated at 70°C for 10 min. dA-tailed cDNA was amplified with primer RACE-TT (5 0 GTTCAGCG-CAGGGTCTTTTTTTTTTTTTT 3 0 ) and the appropriate target-specific primers SC10, SC143 or SC199 (S1 Table). Final amplification was conducted with forward primer RACE-T-F and a nested primer for each reaction (SC10-NEST, SC143-NEST, or SC199-NEST) (S1 Table). Final amplification products were purified and the nucleotide sequence was determined.
mRNA secondary structure prediction The secondary structure in RNAFold (ViennaRNA Package 2.0 [11][12][13]) to Analyses were performed on the full length e 327 bp region from the promoter to the mef(E) start codon and on systematically smaller fragments thereof. In silico, 5' and 3' nested deletions and other fragments were tested to account for the effects of 5' folding during transcription, 5' to 3' degradation, and ribosomal interference.

β-galactosidase assays
To measure expression of mef(E)-lacZ, cultures of XZ7042 and mutant derivatives were grown to mid-log phase (OD 600~0 .3-0.4), divided and exposed to concentrations of erythromycin equal to one-tenth the non-induced MIC of each strain, as determined by Etest (described above). Each subculture was harvested one hour after induction for assessment of β-galactosidase activity. Cultures for time course expression assays were treated as above with cells harvested from continuous cultures at the point of induction and at 30 min intervals thereafter. Expression was determined by standard β-galactosidase protocols [36]. Data was expressed as mean Miller Units ± standard error of the mean (SEM). All experiments were performed in duplicate.

RNA-Seq
The erythromycin (ERY) minimum inhibitory concentrations were determined by microdilution and Etest. Induced MIC was determined in the same manner after overnight growth of the isolates on TSA+ sheep blood agar amended with 10 -1 the erythromycin MIC of each strain. Mid-log cultures of GA17457 were divided equally into four aliquots and each parallel culture was treated with erythromycin (1.2 μg ml -1 ), LL-37 (200 μg ml -1 ), spiramycin (0.1 μg ml -1 ) or left untreated. Cells were harvested after a one hour incubation and total RNA was extracted, enriched for mRNA, reverse transcribed and sequenced.] Strains were induced at midlog growth phase (OD 600 ) as indicated and cells were harvested after one hour incubation. Total RNA from bacterial samples in each condition was extracted using the QIAGEN RNeasy Mini Kit (Qiagen Sciences, Germantown, MD) followed by DNase treatment in solution and rRNA depletion with the Epicentre Ribo-Zero kit (Epicentre, Madison, WI). Strand-specific RNA-Seq cDNA libraries were constructed and sequenced on the Illumina HiSeq2000 platform; two libraries were combined per channel. The data discussed in this publication have been deposited in NCBI's Gene Expression Omnibus [37] and are accessible through GEO Series accession number GSE54176 (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc= GSE54176).

Results
Identification of the transcriptional start site of the mef(E)/mel operon The serotype 19A invasive pneumococcal isolate GA17457 expresses inducible mef(E)/mel-mediated macrolide resistance (Table 1) [8,9]. The start site of the erythromycin-induced mef(E)/ mel transcript from GA17457 was mapped by 5 0 RACE. Unless otherwise stated, all isolates and mutants were induced by one hour exposure to erythromycin at a concentration equal to 10 -1 minimum inhibitory concentration (MIC). The 5' RACE reactions produced an approximately 0.4 kb 5 0 RACE PCR product. The nucleotide sequence of the 5' terminus correlated to a guanine residue located 327 bp upstream of mef(E) and was tagged with a run of 8-10 consecutive thymine residues which is a hallmark of 5' RACE ( Fig. 1A). Consensus-35 (5 0 -TTGAAC-3 0 ) and-10 (5 0 -TATACT-3 0 ) sequences were identified beginning eight base pairs 5 0 of guanine transcriptional start (+1G) and were identical to previously identified pneumococcal promoters (Fig. 1B) [10]. The-10 site was also a consensus "extended-10" pneumococcal promoter (5 0 -TTGTGTTATACT-3 0 ) (Fig. 1B). A second poly-T tract was noted on the 5' RACE chromatograph (Fig. 1A). The second detected transcript appeared to begin at the cytosine residue (+41C) positioned 287 bases upstream of mef(E) (Fig. 1A). However, no promoter sequences were detected immediately upstream of +41C suggesting that it was not an alternate transcription start site. The mRNA molecule terminating with +41C detected by 5' RACE may be a product of transcript degradation and may reflect biologically relevant endoribonuclease activity (see below).
To show that the predicted promoter was required for expression of mef(E), a 37 bp fragment [Δ-41-(-5)] containing the putative-10 and-35 boxes was deleted in the reporter locus inserted into bgaA of XZ7042 (Fig. 1B, 1C). The resulting mutant, XB05, was tested for noninduced and erythromycin-induced expression of mef(E)-lacZ (Fig. 2). As anticipated, the expression of mef(E)-lacZ in XZ7042 containing the wild type promoter was induced rapidly by erythromycin and reached maximum induction (6.8-fold) after 1 hour. Expression in the promoter deletion mutant XB05 was negligible, regardless of erythromycin induction and was similar to the negative control, a promoterless lacZ, strain XZ7049 ( Fig. 2) These data confirm that the promoter was required for transcription of mef(E)-lacZ. Further, site directed mutagenesis of +1G to a thymine (G+1T) in the reporter locus of the GA17457-derived mef(E)-lacZ transcriptional reporter strain XZ7042 resulted in the mutant reporter strain XB16. Under noninducing conditions β-galactosidase (β-gal) activity of XB16 was not significantly different from XZ7042, however, under erythromycin-inducing conditions, expression by XB16 was 3.9-fold lower, confirming that +1G is required for optimal induction of mef(E)/mel (Table 2). To confirm that the promoter was also required for macrolide resistance, the native mef(E)/mel locus of GA17457 was introduced by allele replacement using the kanamycin resistance marker   aphA-3 inserted immediately upstream of Mega, strain XB29. Erythromycin resistance in XB29 containing the aphA-3 and the wild type mef(E)/mel regulatory region was identical to the wild type strain GA17457 (MIC 12 μg ml -1 ). Induction of mef(E)/mel expression by erythromycin resulted in an erythromycin MIC of 64 μg ml -1 for XB29, similar to the induced resistance in GA17457 (MIC 48 μg ml -1 ) indicating that the aphA-3 marker did not influence macrolide resistance or induction. A promoter deletion mutant, XB30, containing the Δ-41-(-5) mutation was sensitive to erythromycin (MIC 0.094 μg ml -1 ) under non-inducing and inducing conditions. Restoration of the Δ-41-(-5) deletion in mutant XB30R restored non-induced and induced erythromycin resistance to wild type levels (MIC 8 μg ml -1 and 64 μg ml -1 , respectively) ( Table 2). Thus, the mef(E)/mel transcript contains a 327 base leader sequence including the promoter and also nucleotides not translated as part of the mef(E) and mel structural genes that be involved in regulation of the operon.
Predictions of the secondary structure of the mef(E)/mel 5' regulatory mRNA Having defined the start of the mef(E)/mel transcript and the promoter, we sought to identify cis-acting regulatory sequences in the 327 base regulatory leader sequence that may contribute to inducible expression of the mef(E)/mel operon. Four pairs of proximal inverted repeats were identified and predicted by RNAFold [11][12][13]) to form four stem-loops (R1/R2, R3/R4, R5/R6 and R7/R8) ( Fig. 1C; Fig. 3). The 54 base repeat R1 annealed to the 51 base R2 to produce a stem-loop (R1/R2) containing a 44 bp stem (37.5% G+C), a 21 base terminal loop. The predicted minimal free energy (ΔG) of the R1/R2 stem-loop was-39.5 kcal mol -1 (Fig. 3). Immediately adjacent to R1/R2 was stem-loop R3/R4 containing an eight base pair stem with a ΔG of-8.1 kcal mol -1 (Fig. 3). Overlapping the inverted repeats R3 and R4 was a pair of perfect 12 base direct repeats (5 0 UUAUUUAACUAU 3 0 ) separated by four bases (red arrows, Fig. 3). Stem-loop R5/R6, located 39 bp downstream of repeat R4, had the characteristics of a classic Rho-independent transcriptional terminator [14]; the R5/R6 loop was predicted to be stable (ΔG = -18.0 kcal mol -1 ), contained a 55.6% G+C-rich 15 bp stem, and was immediately followed by a run of five uracil residues (Fig. 3). The erm(K) transcriptional attenuator has two adjacent Rho-independent transcriptional terminators [15], raising the possibility that last stem-loop R7/R8 in the mef(E)/mel 5 0 regulatory mRNA was a second terminator. However, R7/R8 did not have the characteristics of a Rho-independent terminator; the stem was not G+C-rich (14.3%), it was not predicted to be thermodynamically stable (ΔG-5.3 kcal mol -1 ) and was not followed by a poly-uracil tract.
Identification of a leader peptide mef(E)L required for mef(E)/mel expression Expression of inducible erm methylase genes occurs when erythromycin-bound ribosomes stall during translation of a small leader peptide encoded upstream of the structural gene on the erm transcript. A search of the 327 base 5' regulatory sequence of the mef(E)/mel transcript revealed a small (eight codon) open reading frame (hereafter called mef(E)L) located 34 bases from +1G and preceded by a consensus ribosomal binding site (5' AGGAGG 3') (hereafter called RBS1) ( Fig. 1C; Fig. 3). The peptide sequence of Mef(E)L (MTASRLR) was similar to predicted and confirmed leader peptides of other macrolide resistance determinants (Table 3). Notably, Mef(E)L was homologous to the Bacillus lichenifornis erm(K) leader peptide Erm(K)L (MTHSMRLRFPTLNQ) and the putative leader peptide Msr(A)L (MTASMRLR) encoded upstream of the macrolide-inducible mel(msr(D)) homolog msr(A) from S. epidermidis (Table 3) [1,15]. The similarity of mef(E)L to erm(K) was of note because erm(K) is one of the few erm- Transcriptional Attenuation of mef(E) in S. pneumoniae type methylase genes known to be regulated by transcriptional attenuation [15] and mef(E)/ mel is transcriptionally regulated [4,8]. The minimal leader peptide sequence required for erythromycin induction of erm(K) has been shown to be MTHSMRLR [15,16], nearly identical to Mef(E)L suggesting that the mef(E) leader peptide may be the site of ribosome stalling in the presence of macrolides. Apart from the similarities of the leader peptide, no discernible homology was noted between the 327 base mef(E)/mel 5 0 regulatory region and the 354 base 5 0 regulatory region of erm(K).
To confirm that translation of the predicted leader peptide mef(E)L was required for mef(E)/ mel expression, the native locus mef(E)L was translationally inactivated by altering the ribosomal binding site from AGGAGG (RBS1) to CCTCCT (RBS1') in the mutant XB37 ( Fig. 1C; Table 2). XB37 was sensitive to erythromycin in inducing and non-inducing conditions indicating that mef(E)L expression was required for mef(E)/mel-mediated resistance ( Table 2). These data were consistent with transcriptional attenuation as a mechanism whereby stalled macrolide-bound ribosomes interfere with stem-loop R1/R2 to allow the formation of an antiattenuator.

Stem-loop R5/R6 is a Rho-independent transcriptional terminator
The stem-loop structure R5/R6 was present in the predicted attenuator but absent from the predicted anti-attenuator structure consistent with the conclusion that R5/R6 represents a Rho-independent transcriptional terminator in non-inducing conditions. To determine if R5/ R6 was required for repression of mef(E)/mel expression in non-inducing conditions a 39 bp deletion (Δ+243-280) containing repeat R6 and the 5 0 end of R7 was introduced into the reporter locus of XZ7042 and into the native Mega locus of GA17457 creating XB03 and XB32, respectively (Fig. 1C). The non-induced expression of mef(E)-lacZ in XB03 (89.0 m.u.) was 6.8fold higher than non-induced expression in XZ7042 and was nearly the same as induced expression in XZ7042 (Table 2). However, expression in the XB03 remained inducible, increasing 1.6-fold to 145.7 m.u. after induction by erythromycin (Table 2). This may suggest an additional mechanism of regulation and induction.

Prediction of the anti-attenuator structure of the mef(E)/mel transcript
To identify the anti-attenuator structure, additional inverted repeats in the 327 base leader sequence that could be involved in alternate folding of the mef(E)/mel transcript were identified. A pair adenine-uracil rich 18 base repeats separated by 174 bases, predicted not to anneal were located in the attenuator structure (dashed arrows, Fig. 3). The first unit of these distal repeats, dIR1, was located at the 3' end of R1, six bases downstream of mef(E)L suggesting availability for pairing when a stalled ribosome disrupts the R1/R2 stem-loop. The second distal repeat unit, dIR2 overlapped the Rho-independent transcriptional terminator and included a poly-U tract such that pairing of the distal inverted repeats dIR1 and dIR2 would disrupt the transcriptional terminator and allow transcription to proceed. Under the presumption that ribosome stalling would prevent folding of the transcript from the start through the mef(E)L coding sequence, the 5 0 terminal 63 bases were excluded from interacting in the in silico folding of the regulatory region The predicted transcriptionally active anti-attenuator structure is shown in Fig. 4. The 5 0 terminal 63 bases were excluded from interacting in the in silico folding of the regulatory region under the presumption that ribosome stalling would prevent folding of the transcript from the start through the mef(E)L coding sequence. The leader sequence +60-+327 was predicted to fold into a stable (ΔG-67 kcal mol -1 ) containing a duplex formed by annealing of dIR1 and dIR2 (Fig. 4). Significantly, the putative Rho-independent terminator R5/R6 was resolved, supporting the predicted identity of the anti-attenuator structure (Fig. 4).
To experimentally test the anti-attenuator structure, a deletion of 18 bases corresponding to dIR1 (Δ+63-80) was introduced into the reporter locus of XZ7042 creating XB13 (Fig. 1C). Folding analyses predicted deletion of dIR1 would destabilized the R1/R2 stem-loop, result in the annealing of R6 and the novel sequence created between the truncated R1 and the repeat R2 (Fig. 5), and prevent the formation of the R5/R6 terminator. This favored formation of the anti-attenuator structure even under non-inducing conditions (Fig. 5). The model of attenuation due to a single Rho-independent transcriptional terminator would predict this mutant to constitutively expression mef(E)/mel and macrolide resistance (Fig. 5). Indeed, in non-inducing conditions, expression of mef(E)-lacZ was 4.8 times higher in XB13 (62.2 m.u.) than in the wild-type reporter XZ7042 (13.0 m.u.), and was not further induced by erythromycin exposure (64.8 m.u.) ( Table 2).
The roles of stem-loop R1/R2 and the leader peptide mef(E)L were further tested by introduction of a second deletion (Δ+54-92) into the reporter locus to create the reporter mutant strain XB12 (Fig. 1C). Like Δ+63-80, this mutation deleted dIR1 but extended upstream to disrupt mef(E)L. Folding analyses of the Δ+54-92 deletion predicted that the transcriptional terminator R5/R6 would form in XB12 (Fig. 5) and would form a R1/R2 stem-loop, albeit with reduced stability. The Δ+54-92 deletion also resulted in a truncation of mef(E)L due to substitution of the arginine codon (AGA) in the eighth position with an opal stop codon (UGA) (Fig. 5). The premature termination of the leader peptide was predicted to prevent ribosome stalling and thus render mef(E)/mel expression uninducible.
Indeed, basal expression of β-gal activity in the Δ+54-92 reporter mutant XB12 was not significantly different from the negative controls XZ7049 or XB05 and, under inducing conditions expression was 4.8-fold lower than wild-type (16.6 m.u. and 80.0 m.u., respectively) indicating a nearly complete loss of induction by macrolides (Table 2). Consistent with this result, the Δ +54-92 native locus mutant XB36 (Fig. 1C) was susceptible to erythromycin regardless of prior erythromycin induction (MICs 0.125 μg ml -1 ) ( Table 2). These data confirm the role of stemloop R1/R2, and the leader peptide mef(E)L, and support the theory that induction of mef(E)/ mel is due to ribosome stalling at mef(E)L in the presence of inducing macrolides.

Stem-loop R3/R4 is also required for the regulation of mef(E)/mel expression
Stem-loop R3/R4 was formed by imperfect annealing of R3 and R4 repeats (Fig. 3). R3/R4 contained a pair of 12 base direct repeats (Fig. 3, red arrows) and could function as a cis-acting operator, influencing mef(E)/mel expression through interaction with trans-acting regulatory factors. To determine if R3/R4 influenced mef(E)/mel expression, 28 bp containing the direct repeats (Δ+162-189) was deleted from the mef(E)-lacZ reporter locus to create reporter strain XB14 (Fig. 1C). Folding analyses predicted the terminator would form even without erythromycin induction (Fig. 6). β-gal activity expressed in XB14 grown with or without erythromycin induction was markedly reduced and was similar to the negative controls XZ7049 or XB05, demonstrating that mef(E)-lacZ was not expressed in an R3/R4 mutant (Table 2). Likewise, XB45 containing the Δ+162-189 deletion in Mega was susceptible to erythromycin with or without induction with erythromycin (MICs 0.75 μg ml -1 ) ( Table 2). These data confirmed the requirement of stem-loop R3/R4 in the regulation of mef(E)/mel expression.

Mef(E)/Mel regulation differs from the attenuation model of erm methylases
The observation that the R5/R6 terminator deletion mutants remained inducible to macrolides suggested the presence of unidentified secondary structures or an additional level of regulation. Deletion of the 5' leader sequence, or disruption of the 5' secondary structures, leads to constitutive expression or erm methylases such as erm(A), erm(B) and erm(C) [17,18]. To assess additional levels of regulation, bases +19 to +298 were deleted from the native mef(E)/mel locus and erythromycin resistance was determined. Erythromycin resistance of non-induced the Δ+19-298 mutant, XB31 (Fig. 1C), was not different from the non-induced control strain XB29 (MIC, 12 μg ml -1 ) ( Table 2). However, resistance was not induced significantly by erythromycin induction (MIC, 16 μg ml -1 ) which represented a significant change from the induced control (MIC, 64 μg ml -1 ) ( Table 2). This result was also not consistent with the simple attenuation model observed with erm methylases. Ribosomal stalling protects nascent transcripts from degradation by ribonucleases [19]. The deletion of XB31 removed mef(E)L and thus ribosome stalling was not possible, suggesting that the XB31 mef(E)/mel transcript, though not attenuated, was unstable and quickly degraded, preventing over-expression of the efflux pump. To test this hypothesis, mutant XB38 was generated by deletion of bases +63-298. The mutation was identical to that found in XB31 except that the leader peptide and the putative ribosome stall site remained (Fig. 1C). The noninduced resistance of XB38 to erythromycin (MIC, 32 μg ml -1 ) was 2.7-fold higher than the control XB29, and at least 4-fold higher after exposure to erythromycin (MIC, >264 μg ml -1 ) ( Table 2). These data indicated that the mef(E)/mel transcript was protected from degradation by ribosomes stalled during translation of the leader peptide. Further, 5' RACE detected transcripts that begin with the +41C (Fig. 1A), located in the third codon of mef(E)L, or five codons from its 3' end (see above). This was consistent with the reported endonuclease cleavage site of the erm(C) leader peptide erm(C)L which was shown to be protected by the stalled ribosome [19]. The data suggest that mef(E)L translating ribosomes stall on the last codon of mef(E)L and protect the transcript from further degradation.

Visualization of mef(E)/mel attenuation by RNA-seq
RNA-Seq whole transcriptome analyses were utilized to further determine if the mef(E)/mel transcript was terminated prematurely in non-inducing conditions. Fig. 7 shows the reads mapping to the mef(E)/mel operon viewed using the Integrative Genomics Viewer (IGV) (Broad Institute, Cambridge, MA). The tracks represent expression on the forward strand (reverse strand expression data not shown) under each experimental treatment. As expected, the mef(E)/mel operon was not expressed in untreated samples or in samples treated with the non-inducing macrolide spiramycin (Fig. 7). LL-37, a weak inducer of mef(E)/mel [9], also did not induce mef(E)/mel under these conditions. In the erythromycin-induced sample, mef(E) and mel were expressed as a single operon (Fig. 7). Interestingly, in the non-induced samples, a low level of expression was observed originating downstream of the transcriptional start site and terminating after approximately 250 bases. In the erythromycin-induced sample, Transcriptional Attenuation of mef(E) in S. pneumoniae transcription began at the +1G and continued until the end of the mef(E)/mel operon (Fig. 7). Similar expression patterns for mefE/mel expression (non-induced or induced with erythromycin) were observed in whole transcriptome analyses of three other Mega-containing invasive pneumococcal isolates, GA17545 (AFGA00000000.1), GA02254 (AIKI00000000.1) and GA41565 (AGPO00000000.1) (data not shown). RNA-Seq demonstrated transcriptional attenuation of mef(E)/mel in non-inducing conditions and transcriptional activation in the presence of the inducing macrolide.

Phylogenetic analyses of mef-containing mobile elements
As part of a comparative genomic study of the evolution and dissemination of antimicrobial resistance in a geographically defined population of S. pneumoniae, the genomes of 147 pneumococcal isolates, including 115 macrolide resistant invasive pneumococcal strains, were sequenced. Isolate metadata and the NCBI accession number for each genome are provided in S2 Table. Mega carrying mef(E) was present in 102 macrolide resistant isolates, including 20 that also contained erm(B) carried on Tn2010. Strain England 14 -9 (accession no. NZ_AILI00000000 contained mef(A)/msr(D) located on the mobile element Tn1207.1. All other isolates contained either erm(B) alone (n = 11) or 23S ribosomal mutations (n = 1). Type I Mega (5.5 kb) was found in 61 isolates, 49 isolates had the type 2 Mega (5.4 kb) [3] and a single isolate contained a 112 bp duplication of the 5 0 leader sequence of mel that has not been previously described (data not shown). Of the 102 Mega sequences, 77 were identical and the remaining contained no more than three single nucleotide polymorphisms (SNPs). The exception was the original Mega sequence (AJ274302), which contained a 16 bp insertion that has been described in S. viridians (EF042094) and many mef-containing elements from nonpneumococcal streptococci.
Mef(E)/Mel-mediated resistance to erythromycin varied from 1 μg ml -1 to >64 μg ml -1 in the 102 Mega strains. However, phylogenetic analyses of the Mega sequence revealed no changes within the mef(E) or mel structural genes that correlated with the strain to strain variation in macrolide resistance, suggesting that variable resistance levels were due to differential expression of mef(E)/mel. Deletions in the mef(E)/mel 5 0 regulatory region have not been reported and none were detected in the 102 Mega-containing isolates examined in this study.
Alignment of Mega with pneumococcal elements containing mef(A) (Tn1207.1) and mef(I) (5216IQ complex) using Clustal Omega (version 1.2.1) [20] revealed a 3.2 kb region of homology, "the mef cassette" including the 5 0 regulatory region of mef(E) and mel (Fig. 1A). The mef cassette contained the-10 consensus sequence at its left junction and was inserted in all of the elements just downstream of the-35 box of the mef/mel(msr(D)) operons resulting in evolutionary divergence of the-10 and-35 promoter sequences (Fig. 1A). The promoters displayed some nucleotide similarity (approximately 65% identity), but the 327 bp 5 0 regions of the mef(A) and mef(I) elements were 96% and 97% identical, respectively, to that found in mef(E) (Fig. 1A). These data suggest an evolutionary history of inter-and intra-species horizontal transfer of the 3.2 Kb mef cassette.
A consensus attenuator structure was generated by Clustal Omega alignment of the 5 0 regulatory mRNA region of these other mef-containing elements and prediction of consensus secondary structures performed by RNAlifold (ViennaRNA Package 2.0 [11]). The mefcontaining elements included the pneumococcal mef(I)-and mef(A)-containing elements (5216IQ complex and Tn1207.1, respectively), and elements from S. pyogenes, Clostridium kluyverii, C. perfringens and Group G streptococci. The consensus structure was not significantly different from the mef(E)/mel attenuator structure (Fig. 3).
Further, the sequences of the homologous Rho-independent terminator from these elements were analyzed. The sequence for the predicted terminators of Mega, and Mega carried on Tn2009 and Tn2010 were identical. Other elements displayed one to four SNPs in the predicted terminator (Fig. 8). The mef(I) elements from pneumococcus and S. dysgalactiae subspecies equismilis contained two SNPs (SNPs 1 and 2) (Fig. 8). However, these SNPs correlated to nucleotides of the mef(E) terminator that were predicted to be unpaired and to have minimal impact on the stability of the structure. The mef(A) elements and the mef(I) element from the C. kluyverii phage element contained SNP1 and SNP2, and also displayed SNP3 and/or SNP4 at nucleotides that were predicted to base pair in the mef(E) terminator (Fig. 8). Both SNP3 and SNP4 were adenine substitutions of uracil, thus disrupting an A-T base pair and thus were predicted to result in decreased stability of the structures compared to Mega (Fig. 8).
The overall sequence conservation of the predicted Rho-independent transcriptional terminators suggested similar mechanisms of regulation the mef and mel(msr(D)) genes in each of these elements. In support of this observation, the erythromycin MIC of strain England 14 -9 containing mef(A) on Tn1207.1, was induced 8-fold by subinhibitory concentrations (1.0 μg ml -1 ) of erythromycin, increasing from 20 μg ml -1 in the absence of erythromycin to 320 μg ml -1 when induced. These data suggest that transcriptional attenuation is the mechanism of repression/induction of efflux-mediated macrolide resistance in Gram-positive bacteria regardless of the mef allele and its associated mobile element.

Discussion
In this study the mechanism of repression and induction of the macrolide efflux operon mef (E)/mel encoded on the Mega mobile element in S. pneumoniae and other Gram-positive bacteria was defined. Like erm(B), mef(E)/mel is induced by the 14-and 15-membered macrolides such as erythromycin and azithromycin, but not by 16-membered macrolides such as spiramycin and midecamycin [21]. This suggested that the two-systems were controlled by similar mechanisms, that is, attenuation.
Attenuation of inducible erm genes typically occurs at the level of translation (reviewed in [7]) and is characterized by the presence of one or two small (8-20 amino acids) leader peptides located on the erm transcript 5' of the start codon encoding the structural Erm protein.
These leader peptides toggle the folding of the erm transcript between two alternate secondary structures; the attenuator that blocks expression of erm, and the anti-attenuator that promotes expression. Ribosomes bound by an inducing macrolide pause at a programmed stall site on the leader peptide(s) and disrupt the attenuator, allowing formation of the anti-attenuator. Constitutively expressed erm genes typically have mutations in their 5' regulatory regions which prevent formation of the attenuator structure [21]. In some erm macrolide-inducible systems, attenuation occurs at the level of transcription. This is similar to translation attenuation with the distinction that transcriptional attenuators form Rho-independent transcriptional terminators in non-inducing conditions leading to premature termination of transcription of the structural erm gene. Transcriptional attenuation has been observed for erm-encoded methylases of the erm(D) class, which includes erm(D), erm(K) and erm(J) found in Bacillus species [15,16,22]. We have previously found that expression of mef(E)/mel is controlled at the level of transcription [4,8]. Thus, the hypothesis that mef(E)/mel was transcriptionally attenuated was explored.
The transcriptional start (+1) of the mef(E)/mel operon was mapped by 5 0 RACE to a guanine nucleotide located 327 bp upstream of the mef(E) start codon. A consensus pneumococcal promoter sequence was identified upstream that included a pneumococcal extended-10 box and a-35 box. Deletion of the putative promoter abolished expression of a mef(E)-lacZ transcriptional fusion and rendered the mutant susceptible to erythromycin. Restoration of the mutated promoter to the wild type restored resistance to wild type levels confirming the location of promoter.
Analyses of the 327 bp 5 0 regulatory region of mef(E) downstream of the promoter revealed a feature common to attenuators involved in the regulation of erm-type methylases, including a small orf, mef(E)L, encoding a small peptide (MTASMRLR) ( Table 2) preceded by a consensus ribosomal binding site (RBS1). The molecular signals dictating programmed ribosome stalling involve sequence-dependent interaction between the nascent peptide, the peptidyl transferase center in the exit tunnel and the inducing macrolide molecule [23]. Macrolideinduced stalling at sites located on the leader peptides prevents the attenuator from forming in favor of the anti-attenuator to promote expression of the Erm methylase. The amino acid sequence of Mef(E)L was very similar to the leader peptide identified in other inducible macrolide systems, including Msr(A)L (MTASMRLK) encoded upstream of msr(A) Staphylococcus and the transcriptionally attenuated erm(K) (MTHAMRLRFPTLNQ) and erm(D) (MTHSMRLR) genes [15,16]. Significantly, the stall site on the erm(K) leader peptide erm(K)L is MRLR and the minimal Erm(K)L sequence that will allow induction of erm(K) is MTHAMRLR. These are nearly identical to the predicted sequence of the Mef(E)L peptide. We demonstrated that mef(E)L leader peptide was required for induced transcription of the mef (E)/mel operon and for efflux-mediated macrolide resistance. Translational inactivation by of mef(E)L ribosomal binding site by site directed mutagenesis and truncation of the protein resulted in macrolide susceptibility regardless of induction, consistent with the requirement for ribosomal stalling and anti-attenuation.
Transcriptional attenuators require one or more Rho-independent terminators 5' of the induced structural gene. Stem-loop R5/R6 (Fig. 6) was identified as the terminator of the mef(E)/ mel attenuator. It has a G+C rich stem, was predicted to be stable, and was immediately followed by a run of five uracil residues. Deletion of the structure significantly increased basallevel expression and resistance to erythromycin. Stem-loop R3/R4 was also required for the regulation of mef(E)/mel expression and we identified a pair of distal 17 base perfect inverted repeats in the mef(E) 5 0 regulatory region separated by 174 nucleotides. The location of the inverted repeats within critical structures of the attenuator suggested involvement in the attenuation mechanism. Deletion of the first inverted repeat sequence, IR1, resulted in high-level expression, but abolished induction by erythromycin confirming a role in mef(E)/ mel regulation.
The RNA-Seq data provided additional clarity and strong evidence for a transcriptional attenuation model and the induction of mef(E)/mel by anti-attenuation of transcription in the presence of inducing macrolides. In untreated and non-macrolide antibiotic-treated samples, the sequence reads mapped precisely to the transcriptional start site and terminated at the 3 0 end of the Rho-independent terminator. In the macrolide-induced sample, the reads mapped to the start site and continued through mef(E) and mel structural genes. Our observations do suggest that control of mef(E)/mel is influenced by regulatory mechanisms in addition to transcriptional attenuation. We have previously reported that the antimicrobial peptide LL-37 induces mef(E) expression and increased resistance to erythromycin and that the mechanism of induction appeared to be distinct from that of macrolides [9]. Also, in some attenuator deletion mutants expression and resistance were still inducible by erythromycin indicating an additional level of control of the mef(E)/mel promoter. Further, multiple mechanisms have been reported controlling erm gene expression. In addition to attenuation, erythromycin-induced ribosomal stalling increases expression of erm genes by protecting transcripts from 5 0 -to-3 0 nucleolytic degradation [24][25][26]. In preliminary experiments we have observed that in mef(E)/ mel attenuator-deficient mutants, erythromycin resistance was reduced when mef(E)L was deleted as compared to when mef(E)L and the ribosome stall site were present. This suggests that erythromycin-mediated ribosome stalling may increase mef(E)/mel mRNA stability in a similar manner as reported for several erm genes. Another regulatory mechanism has recently been reported for erm(C) that involved macrolide-induced ribosomal frameshifting [27]. Frameshifting during translation of the leader peptide erm(C)L requires an telithromycin-bound ribosome and an erm(C)L"shift-prone" sequence which is essentially a run of four or more uracil or adenine residues. The ribosomal shift promotes translation of erm(C) by a mechanism that has yet to be determined. We have previously demonstrated that telithromycin induced mef(E)/mel expression and increased resistance to non-ketolide antibiotics [8]. The other potential regulatory mechanisms will be explored in future work.
There are other important differences between erm methylase and mef(E)/mel regulation suggesting that the systems are not completely analogous. We have shown previously that induction of mef(E)/mel by macrolides is dependent upon the identity of the amino sugar side chains of the macrolide ring [21]. Additionally, constitutively resistant (cMLS B phenotype) clinical isolates of many species have been identified and are readily selected in the laboratory setting ( [17,28,29]. To date, there have been no reports of constitutively expressed mefmediated macrolide efflux in clinical isolates or generated in vitro. This may suggest a severe fitness cost to unregulated expression of mef(E)/mel in contrast to the erm methylases.
The comparisons of the 5 0 region of mef(E)/mel of Mega to mef-containing elements from pneumococci and other Gram-positive species revealed conservation of the 327 bp regulatory region and the predicted secondary attenuator structures. These data suggest that mef(E), mef (A) and mef(I) genes and the adjacent msr genes in pneumococci, other streptococci and other Gram-positive species, are likely responsive to macrolide induction and regulation by similar transcriptional attenuation. We verified that mef(A)/msr(D) carried on Tn1207.1 was induced by erythromycin. Thus, mef(E)/mel in S. pneumoniae is a model for inducible mef-mediated expression in a variety of Gram-positive pathogens.
Supporting Information S1