Beyond Self-Resistance: ABCF ATPase LmrC Is a Signal-Transducing Component of an Antibiotic-Driven Signaling Cascade Accelerating the Onset of Lincomycin Biosynthesis

ABSTRACT In natural environments, antibiotics are important means of interspecies competition. At subinhibitory concentrations, they act as cues or signals inducing antibiotic production; however, our knowledge of well-documented antibiotic-based sensing systems is limited. Here, for the soil actinobacterium Streptomyces lincolnensis, we describe a fundamentally new ribosome-mediated signaling cascade that accelerates the onset of lincomycin production in response to an external ribosome-targeting antibiotic to synchronize antibiotic production within the population. The entire cascade is encoded in the lincomycin biosynthetic gene cluster (BGC) and consists of three lincomycin resistance proteins in addition to the transcriptional regulator LmbU: a lincomycin transporter (LmrA), a 23S rRNA methyltransferase (LmrB), both of which confer high resistance, and an ATP-binding cassette family F (ABCF) ATPase, LmrC, which confers only moderate resistance but is essential for antibiotic-induced signal transduction. Specifically, antibiotic sensing occurs via ribosome-mediated attenuation, which activates LmrC production in response to lincosamide, streptogramin A, or pleuromutilin antibiotics. Then, ATPase activity of the ribosome-associated LmrC triggers the transcription of lmbU and consequently the expression of lincomycin BGC. Finally, the production of LmrC is downregulated by LmrA and LmrB, which reduces the amount of ribosome-bound antibiotic and thus fine-tunes the cascade. We propose that analogous ABCF-mediated signaling systems are relatively common because many ribosome-targeting antibiotic BGCs encode an ABCF protein accompanied by additional resistance protein(s) and transcriptional regulators. Moreover, we revealed that three of the eight coproduced ABCF proteins of S. lincolnensis are clindamycin responsive, suggesting that the ABCF-mediated antibiotic signaling may be a widely utilized tool for chemical communication.

IMPORTANCE Resistance proteins are perceived as mechanisms protecting bacteria from the inhibitory effect of their produced antibiotics or antibiotics from competitors. Here, we report that antibiotic resistance proteins regulate lincomycin biosynthesis in response to subinhibitory concentrations of antibiotics. In particular, we show the dual character of the ABCF ATPase LmrC, which confers antibiotic resistance and simultaneously transduces a signal from ribosome-bound antibiotics to gene expression, where the 59 untranslated sequence upstream of its encoding gene functions as a primary antibiotic sensor. ABCF-mediated antibiotic signaling can in principle function not only in the induction of antibiotic biosynthesis but also in selective gene expression in response to any small molecules targeting the 50S ribosomal subunit, including clinically important antibiotics, to mediate intercellular antibiotic signaling and stress response induction. Moreover, the resistance-regulatory BGC mediated by an ABCF resistance protein. At the same time, the dual resistance-regulatory function of the LmrC ABCF protein reported here is unprecedented.

RESULTS
LmrC antibiotic resistance protein is dispensable for resistance. To interpret the role of the three resistance proteins encoded in the lincomycin BGC, we first evaluated the contribution of the individual proteins to the resistance. Specifically, we knocked out lmrA, lmrB, and lmrC singly or in pairs in the lincomycin-producing S. lincolnensis wild-type (WT) strain (Fig. 1b) and, in addition, we complemented the genes under the The MIC values are given as the means 6 the SD of n $ 3 independent measurements. The level of significance of the fold change relative to WT (upper graph) or M1154 (lower graph) is shown (*, P , 0.05; **, P , 0.01; ***, P , 0.001). The "." sign points to values that exceed 16,384 mg liter 21 . The levels of lincosamide production for S. lincolnensis (n = 8) and S. caelestis (n = 10) are given as a percentage of the maximum production achieved by each strain. The overall table of lincosamide susceptibilities of complemented knockout mutants and MIC values for the mycelia of different growth stages are available in Fig. S1. control of a constitutive or natural promoter acting in trans (see Fig. S1). Furthermore, we constitutively expressed the genes in a lincosamide-sensitive Streptomyces coelicolor M1154 strain (26) (Fig. 1c). Then, we evaluated the resistance phenotype of the WT, knockout, and complemented strains by determining the MICs of lincomycin and its derivative, clindamycin.
We revealed that all strains bearing lmrA, including the strains with lmrA only (Fig. 1b, S. lincolnensis strains WT, DC, DB, and DBC; Fig. 1c, S. coelicolor strains AC C and A C ), were highly or extremely resistant to lincomycin, while the resistance to lincomycin significantly decreased when lmrA was absent regardless of other two resistance genes were present. Interestingly, the resistance to clindamycin, generally a more efficient semisynthetic derivative of lincomycin, is different in this respect. Specifically, the majority of the tested strains were moderately resistant to clindamycin with no or little contribution of LmrA to the resistance (Fig. 1b, compares the strains differing in lmrA only: DB versus DAB, WT versus DA, and DC versus DAC, and Fig. 1c, A C versus M11541pIJ10257). Therefore, we assume that LmrA, a transporter of the major facilitator family, is highly specific to lincomycin but not clindamycin, and it ensures sufficient self-resistance to the produced lincomycin on its own.
In contrast to the LmrA transporter, the LmrB 23S rRNA monomethyltransferase confers high resistance to lincomycin and clindamycin when overexpressed in S. coelicolor The last resistance protein, LmrC, confers moderate resistance to both lincomycin and clindamycin when overexpressed in S. coelicolor (Fig. 1c, strain C C ). However, its contribution to the overall resistance in S. lincolnensis is not considerable relative to either LmrB (Fig. 1b, compares strains DA versus DAC) or LmrA (Fig. 1b, compares strains DB versus DBC). The DC knockout strain without lmrC showed a slightly increased lincosamide resistance compared to that of the WT (Fig. 1b).
It is worth noting that the complementation of the knockout strains under the control of the putative natural promoter restored the resistance phenotype of the WT except for lmrB (see Fig. S1 in the supplemental material). In this case, complementation had to be performed under the control of a constitutive promoter because lmrB is cotranscribed with three upstream genes, as evidenced below. Interestingly, constitutive lmrB expression resulted in higher resistance values than those of lmrB expression in its original genomic context (see Fig. S1). Furthermore, the resistance of S. lincolnensis WT and knockout strains was determined from spore suspension, which does necessarily reflect the resistance of the mycelium during lincomycin production. Therefore, we determined the MICs of S. lincolnensis deletion strains using spores and mycelia from two different time points of the seed or production cultures (see Fig. S1). Overall, the data for mycelia comply with the data obtained for spores and show that resistance of the mycelium during production increased compared to the mycelium from the seed culture.
Apart from investigating the resistance phenotypes, we determined the amount of lincomycin produced by S. lincolnensis WT and single-knockout strains in the culture broth (Fig. 1b). The results support our conclusions drawn from the resistance of the strains. Specifically, the strains with high or extreme resistance to lincomycin were able to produce considerable levels of lincomycin, i.e., the strains bearing both lmrA and lmrB (which had the largest amount of lincomycin produced) and the strain bearing lmrA and not lmrB (which had up to 50% of the largest amount of lincomycin produced). On the other hand, strains without lmrA, which were the least resistant to lincomycin, produced only traces of lincomycin or nothing.
The dispensability of LmrC for the overall resistance documented above complies with the comparable lincomycin production of DC versus WT strains. In addition, the production of lincomycin significantly fluctuated (Fig. 1b). This observation could be explained by a more complex regulation-resistance system (LmbU, LmrA, LmrB, and LmrC) encoded within the lincomycin BGC compared to the highly similar BGC of another lincosamide, celesticetin (12), which contains only one nonbiosynthetic gene, the ccr1, coding for Ccr1 23S rRNA monomethyltransferase, as a self-protecting resistance protein homologous to LmrB (Fig. 1a) (for a review, see reference 27). Indeed, the fluctuation of celesticetin produced by S. caelestis in parallel cultures is not as pronounced as that of the lincomycin produced by S. lincolnensis. Moreover, S. caelestis never failed to produce celesticetin (Fig. 1d), while S. lincolnensis failed to produce lincomycin in several parallel WT cultures (Fig. 1b).
Expression of lmrA, lmrB, lmrC, and lmbU is induced by clindamycin. Given our hypothesis of the complex regulation-resistance system of lincomycin production, we wondered whether the expression of any of the lmbU, lmrA, lmrB, and lmrC genes could be affected by the produced antibiotic. Therefore, we cultured S. lincolnensis WT and divided the culture before the onset of lincomycin biosynthesis into two parallel cultures, one of which was supplemented with clindamycin at a subinhibitory concentration ( Fig. 2a). At several time points, we semiquantitatively monitored the expression of the respective genes by reverse transcription-PCR (RT-PCR). Supplementation with clindamycin allowed us to distinguish between the lincosamide used to study its effect on the gene expression and the lincosamide produced by the strain, which we determined by ultrahigh-performance liquid chromatography.
The results in Fig. 2a show that supplementation with clindamycin induced the expression of all the studied genes-lmbU, lmrB, lmrA, and lmrC-in the earlier stages of growth (9 to 16 h) compared to that of the untreated cultures (44 to 104 h), while no effect of clindamycin was observed on the rpoD control. In agreement with this observation, the onset of lincomycin production also shifted toward an earlier time of cultivation (44 h) in the cultures supplemented with clindamycin and reached higher ). The transcription of lmrA was induced more readily and was detectable over a longer period, while the relative amount of lmrC, lmbU, and lmrB transcripts decreased over time. Similar profiles of lmrC, lmbU, and lmrB transcripts indicate that these genes might be in the same operon. Amplification of lmbU from the 1st DNA strand synthesized using a primer specific to lmrB demonstrated that the expression of lmrB is directly coupled with that of lmbU and the two biosynthetic genes lmbX and lmbY (Fig. 2c). On the other hand, an analogous mapping of the start of lmbU transcript (Fig. 2d) and the end of lmrC transcript (see Fig. S3a) showed that the lmrC gene is transcribed independently of the lmbUXY-lmrB operon.
LmrC is essential for the antibiotic-induced onset of lincomycin production. Given the newly defined function of ABCF proteins as modulators of ribosomal PTC, the onset of lincomycin production in response to antibiotics might be regulated by LmrC. To uncover the role of LmrC, we performed comparative mass spectrometry proteomic analysis of the mycelia of S. lincolnensis WT, WT1C c, and DC strains grown in the absence or presence of clindamycin. As shown in Fig. 3a, clindamycin supplementation increased the abundance of lincomycin BGC proteins in both the WT (namely, proteins of the lmbUXYB operon) and WT1C c (the whole BGC), while in the DC strain, lincomycin BGC proteins were more abundant in cultures without clindamycin. The induction by clindamycin was also observable at the lincomycin production level at 40 h in the WT1C c strain but not in the DC knockout strain, where higher production levels were independent of clindamycin treatment (Fig. 3a). No lincomycin was detected in clindamycin supplemented WT cultures, which contradicts the experiment in Fig. 2a, where high levels of lincomycin were detected at the end of seed culture supplemented with clindamycin. Slight differences in cultivation conditions might be responsible for the shifted onset of lincomycin production between the two experiments (see Fig. S7b). Nevertheless, these results suggest that LmrC is required for the onset of lincomycin biosynthesis triggered by clindamycin. To confirm that LmrC is essential for the transduction of antibiotic signal to the expression of lincomycin BGC, we quantified the transcripts of lmrC, lmbU, and lmbN genes in S. lincolnensis WT and DC cultured with or without clindamycin at a time point before lincomycin BGC expression. As shown in Fig. 3b, the clindamycin induced transcription of lmrC, lmbU, and lmbN, which was not under the direct control of lmbU, in the WT strain but not in the lmrC-deficient DC knockout strain. Notably, the observed low-level constitutive transcription of lmbU in the DC strain can be explained by the insertion of apramycin cassette (see Fig. S3b), causing a polar effect. This phenomenon explains the increased production of proteins in DC (Fig. 3a). However, it is important for our reasoning that neither protein production nor lmbU transcription in the DC strain is affected by clindamycin. ABCF family proteins generally exhibit ATPase activity, which is required for protein function (22). Therefore, we wondered whether LmrC is a functional protein capable of ATPase activity that induces gene expression. Hence, we complemented the DC knockout strain with lmrC or lmrC EQ12 expressed from a theophylline-inducible plasmid (C i ). The overproduction of functional LmrC resulted in the expression of lmbU and lmbN, while the overproduction of ATPase-deficient LmrC EQ12 mutant did not have this effect (Fig. 3c). Notably, the expression of lmbU and lmbN mediated by the overproduction of LmrC was achieved without supplementation with clindamycin, and a similar phenomenon was observed at the protein level when LmrC was produced constitutively in the WT (the comparison of WT and WT1C C without clindamycin treatment is shown in Fig. 3a). Altogether, these results demonstrate that clindamycin induces the production of LmrC, which in turn induces the production of LmbU, which is a known activator of lincomycin biosynthesis (24).
The antibiotic-LmrC-LmbU signaling cascade is independent of other S. lincolnensis regulatory elements. Several recent studies described regulators of lincomycin biosynthesis encoded outside the BGC in the S. lincolnensis genome (28)(29)(30)(31). Some of these conserved global regulators might be involved in the antibiotic-induced onset of lincomycin production in addition to LmrC. To rule out this hypothesis, we cloned the lincomycin BGC region starting upstream of lmrC and ending with lmbU translationally fused with the mCherry reporter (C-U-mCh) and introduced it into S. coelicolor M145. Truncated versions of the construct without the 59-half lmrC (U-mCh) and lmbU-mCherry expressed from the constitutive ermEp promoter (U C -mCh) were used as controls (Fig. 3d). As expected, the induced production of the mCherry reporter in response to clindamycin was detected only in the strain with the full-length C-U-mCh construct, while in the strain with the truncated lmrC gene (U-mCh), the level of mCherry expression did not change after clindamycin induction. From these data, we concluded that lmbU expression, and thus the onset of lincomycin production in response to antibiotics, is mainly triggered by lmrC. However, we cannot completely exclude the possibility that S. coelicolor homologs of global regulators that contribute to lincomycin biosynthesis in S. lincolnensis, such as AdpA and BldD (32,33), also affect the expression of lmrC and lmbU in S. coelicolor.
LmrC production is induced by the LS A P group of antibiotics and dampened by LmrA and LmrB. We have shown that the LmrC-induced transcription of lmbU is largely dependent on how LmrC production is regulated in response to antibiotics. To , and lmrC knockout (DC) strain were cultured for 40 h in seed medium with or without clindamycin (CLI) and for another 120 h in production medium without clindamycin. Lincomycin production was quantified after 40 h and at the end of production culture (160 h). Values of lincomycin production are independently expressed as a percentage of maximum production at each time point. Statistical analysis of the proteomic data is shown in Fig. S4a and Table S1. A comparison of WT growth with or without clindamycin supplementation is available in Fig. S4b. Resistance genes (in red) are indicated by the prefix "r." The orientation of genes in lincomycin BGC are indicated by "less-than" (,) and "more-than" (.) signs. (b and c) Results of the qRT-PCR analysis (n = 4) of the WT and DC from the 16-h seed culture show that LmrC is required to activate lmbU and lmbN transcription in response to clindamycin (b) and that only ATPase-active LmrC can activate lmbU and lmbN transcription (c). The production of LmrC and its ATPasedeficient mutant LmrC EQ12 in DC were inducible by theophylline (DC1C i and DC1C EQ12i , respectively). The data are expressed relative to the WT cultured with clindamycin; the error bars indicate standard deviation, and the asterisks represent the level of significance (*, P , 0.05; **, P , 0.01). The evidence that LmrC EQ12 does not affect growth is shown in Fig. S4c. (d) Western blot showing the induction of lmbU expression by clindamycin-induced LmrC in the heterologous host S. coelicolor M145 with the respective plasmids. U-mCherry (U-mCh) levels in 16-h seed culture mycelium uninduced or induced by clindamycin (0.03 mg liter 21 ) were detected by mCherry-specific antibody in S. coelicolor M145 carrying plasmids C-U-mCh, U-mCh, and U C -mCh. A representative Western blot is shown. The graph below shows the average values of relative protein abundancy (RPA) in three independent experiments. ABCF ATPase LmrC Regulates Lincomycin Biosynthesis ® gain more insight into the regulation of LmrC production, we first tested whether the inducing antibiotics were limited only to clindamycin. For this purpose, we treated S. lincolnensis WT cells with a range of ribosome-targeting antibiotics and a cell wall-targeting carbenicillin, and we detected LmrC protein levels using an LmrC-specific antibody. In addition to clindamycin, only lincomycin (lincosamide group), pristinamycin IIA (streptogramin A group), and tiamulin (pleuromutilin group), with lower efficiency, induced LmrC production (Fig. 4a). Interestingly, the activity of these lincosamidestreptogramin A-pleuromutilin (LS A P) antibiotics with overlapping binding sites on the ribosome (Fig. 4a) is compromised by antibiotic resistance ABCF proteins as exemplified by Vga(A) LC in staphylococci (18,34,35). This suggests that the regulation of LmrC production is coupled to its resistance function, which is the dislocation of the antibiotic from its specific overlapping binding sites within the PTC. Next, we investigated whether the lincomycin BGC-encoded resistance proteins LmrA and LmrB can affect the whole cascade by dampening lmrC expression. First, we evaluated LmrC protein LmrC production is not induced by lincomycin and clindamycin in DB1B C with constitutive lmrB expression because lincosamides cannot bind to the ribosomes methylated by LmrB. In contrast, pristinamycin IIA, which can bind to methylated ribosomes, retained the ability to induce LmrC production. (c) Lincomycin produced in the media of the WT (n = 6), lmrB knockout (DB, n = 2), and DB1B C (n = 3) strains after 42 h of seed culture with or without clindamycin supplementation (see Fig. S7b). (d) LmrC production in response to lincomycin was higher in the lmrA-null mutant (DA) lacking a specific lincomycin exporter than in the WT. In panels a, b, and d, representative Western blots are shown, and the graphs below show the average values of relative protein abundancy (RPA) in three independent experiments. levels in S. lincolnensis DB1B with constitutive overproduction of the LmrB methyltransferase. As shown in Fig. 4b, LmrC production no longer responded to lincomycin and clindamycin, which do not bind to ribosomes methylated by LmrB, but instead remained responsive to the treatment with pristinamycin IIA (streptogramin A group), which can bind to methylated ribosomes (36). Since the lmrB gene is in the lmbUXY-lmrB operon encoding biosynthetic enzymes ( Fig. 2c and d), LmrB is an ideal candidate to provide a feedback loop of the cascade. Indeed, pronounced induction of lincomycin production was apparent in strain DB compared to the WT, whereas in the DB1B strain, constitutive lmrB expression dampened the onset of lincomycin biosynthesis (Fig. 4c).
Next, we evaluated the effect of LmrA on LmrC protein levels. Since LmrA confers high-level resistance only to lincomycin (Fig. 1b; see also Fig. S1), there was a considerably higher level of LmrC in the WT induced by clindamycin than in the WT induced by lincomycin ( Fig. 4a and b), which may reflect the fact that only lincomycin is exported by the LmrA transporter; thus, the low intracellular levels are maintained (37). Indeed, deletion of lmrA (DA) resulted in comparable levels of LmrC expression induced by either lincomycin or clindamycin (Fig. 4d). LmrA thus specifically dampens the LmrC production induced by lincomycin by reducing its intracellular concentration. Since both LmrB and LmrA reduce lmrC expression in response to antibiotics, we propose that alongside their resistance function, they also serve as a negative feedback loop to the antibiotic-LmrC-LmbU signaling cascade of lincomycin biosynthesis.
LmrC production is regulated by ribosome-mediated transcriptional attenuation. Substantially reduced production of LmrC in the lmrB-overexpression strain (Fig. 4b) after antibiotic induction showed that the binding of the antibiotic to the ribosome is a prerequisite for the induction of LmrC production. LmrC could thus be regulated by a ribosome-mediated attenuation mechanism as described previously for other antibiotic-resistant ABCF proteins (38)(39)(40): in the absence of antibiotics, either the formation of a premature terminator in the 59 untranslated region (59UTR) or the inaccessibility of ribosome binding site (RBS) prevent gene expression. In the presence of antibiotics, inhibited ribosomes stall during translation of the upstream regulatory open reading frame (uORF), which promotes the alteration of the 59UTR secondary structure and thereby releases gene expression (41). Indeed, an in silico analysis of the lmrC upstream region revealed two putative promoters and two premature terminators with the ability to form alternative antiterminator conformations and several short uORFs (Fig. 5a). To examine whether the attenuation mechanism is involved in the control of lmrC expression, we first used RT-PCR to map from which of the two predicted promoters lmrC is transcribed and the position of the premature terminator (Fig. 5a). The analysis of RNA from the 16-h time point, where lmrC is induced by clindamycin, and from the 104-h time point, where lmrC transcription starts naturally without clindamycin supplementation (Fig. 2a), showed that in both cases, the lmrC transcript starts from promoter P1 (Fig. 5b). As shown in Fig. 5c, the position of the premature terminator was mapped to the region between primers RT 4 and 5, which corresponds to the position of the predicted terminator 1 (Fig. 5a). Next, we prepared a reporter system in which the lmrC upstream region, including its promoter, and full-length lmrC were translationally fused to mCherry (C-mCh). We introduced the construct into S. lincolnensis WT and S. coelicolor M145 strains and determined mCherry levels with or without clindamycin (see Fig. S5d). The mCherry-specific signal was detected only in the presence of clindamycin in both strains, so further experiments were performed in S. coelicolor M145. A series of G-to-C and C-to-G point mutations (see Fig. S6a) in the terminator hairpin led to the disruption and restoration of clindamycin-induced C-mCh production, confirming the terminator prediction (Fig. 5d). To localize the uORF, we mutated the start codons of four upstream ORFs (ATG to ATC or ATG to AAG; Fig. S6a). Surprisingly, only the disruption of uORF2, which partially overlaps with the terminator, led to strong constitutive expression of C-mCh, whereas mutations in other ORFs did not affect C-mCh production (Fig. 5d). This observation suggests an unusual attenuation mechanism in which uORF2 translation is required to form a terminator structure. In summary, the antibiotic-mediated control of lmrC expression occurs via the formation of a premature terminator structure, which prevents lmrC expression in the absence of an antibiotic. LS A P antibiotics, if bound to PTC, trigger the shift from the terminator to the antiterminator conformation, enabling lmrC transcription (see Fig. S6a). The lmrC transcript, specifically its attenuator, is thus the primary sensor of the antibiotic-LmrC-LmbU signaling cascade for lincomycin biosynthesis.
LmrC is coproduced with seven other ABCFs, two of which are responsive to a lincosamide. The LmrC ABCF protein has a regulatory function, which transduces an antibiotic signal to activate lincomycin biosynthesis in S. lincolnensis. Comprehensive phylogenetic analysis classified 30 subfamilies of bacterial ABCF proteins (22). Four subfamilies (Uup, Etta, YdiF, and YbiT) have a broad distribution, while others, including subfamilies with the resistance function (ARE1-7), are taxon specific. Actinobacteria is a phylum with the highest number of ABCFs, including seven subfamilies specific to this taxon (AAF1-6, ARE4-5). In addition to LmrC, which belongs to the ARE5 subfamily, the genome of S. lincolnensis encodes eight ABCF proteins, three of which have putative resistance activity (ARE5 encoded by SLINC_7152 and two AAF4 encoded by SLINC_1109 and SLINC_6197). We speculated whether some of these resistance proteins are induced by clindamycin and thus could have an antibiotic-responsive regulatory function. We used the same mass spectrometry proteomics data set as used for the comparative analysis of lincomycin biosynthetic proteins to analyze the abundance of ABCF proteins in S. lincolnensis WT, WT1C c , and DC strains grown in the absence or presence of clindamycin. As shown in Fig. 3a, all but one ABCF protein was present in all samples, but only two (ARE5 and AAF4) out of three putative antibiotic-resistant ABCF proteins were substantially upregulated by clindamycin or produced lincomycin. The third putative resistance ABCF protein was not detected in any of the samples. Considering that the putative resistance function of these clindamycin-responsive ABCF proteins is redundant, they may have a regulatory function similar to LmrC.

DISCUSSION
Antibiotic resistance proteins associated with BGCs have traditionally been perceived as a means of self-protecting mechanisms. It has been proposed that the expression of multiple resistance genes within the same BGC is regulated to optimize the self-protective resistance levels at different stages of growth or biosynthesis to minimize the fitness cost of the resistance expression (42) or to synchronize the resistance in sibling cells (43,44).
In this study, we characterized an LS A P antibiotic-driven signaling cascade for the activation of the onset of lincomycin biosynthesis, in which an antibiotic resistance protein, LmrC, from the ARE5 subfamily of ABCF proteins is the key signal-transducing element (Fig. 6a). The mechanism lies in the induction of lmrC transcription by ribosome-mediated attenuation, which means that lmrC, specifically its attenuator-forming upstream 59UTR transcript, is a sensor of LS A P antibiotics. Ribosome-mediated attenuation is a common mechanism of regulation of antibiotic resistance ABCF genes in Firmicutes (38- 40,45). However, we describe here for the first time its function as a sensor of the signaling cascade. The major novelty of this cascade lies in the dual antibiotic resistance and regulatory function of the ABCF protein LmrC, which transduces the antibiotic signal to the expression of LmbU and promotes lincomycin biosynthesis. In addition, we show that another two lincomycin BGC-encoded resistance proteins, LmrB and LmrA, affect the cascade by dampening the LS A P antibiotic-induced expression of lmrC. We assume that LmrB, due to its position in the lmbUXY-lmrB operon with two biosynthetic genes, mediates a direct negative feedback loop of the cascade. The LmrA transporter links lincomycin biosynthesis to the primary metabolic pathways since it is regulated by the GlnR global regulator (37). LmrA seems to be the most important component for lincomycin production because lincomycin biosynthesis is remarkably suppressed when LmrA is not present. Furthermore, LmrA, as a lincomycinspecific transporter, desensitizes the cascade specifically to lincomycin, which may prevent the products from reactivation the biosynthesis when it is no longer desirable. In addition, the active export of lincomycin contributes to the propagation of antibiotics within the population.
The last component of the regulation cascade, LmbU, is a transcriptional regulator of the newly proposed LmbU family (24). The lmbU gene has been evolutionarily accepted along with genes encoding the biosynthesis of the unusual precursor 4-alkyl-L-proline (27), which is a building block of lincomycin and other natural products from Streptomyces (46)(47)(48). On the other hand, the LmbU homolog is missing in the closely related BGC for the lincosamide celesticetin, which contains proteinogenic L-proline instead of 4-alkyl-L-proline in its structure.
The regulatory pair of LmbU and LmrC is unique to lincomycin BGC; no other known BGC encodes a LmbU-family regulator together with an ABCF protein. On the other hand, BGC-associated ABCF proteins were almost exclusively present in the BGCs for PTC-targeting antibiotics ( Fig. 6b; see also Table S1). Most of these BGCs encode additional resistance determinants and pathway-specific transcriptional regulators; however, none are homologous to LmbU (see Table S1). We hypothesize that BGCencoded ABCF proteins employ transcriptional regulators of various families to form a signaling cascade to activate the biosynthesis of ribosome-binding antimicrobials.
It was previously shown that LmbU directly activates only the 4-alkyl-L-proline biosynthesis-encoding part of lincomycin BGC (24), which is also evident from our proteomic data (Fig. 3a). In contrast, the recently described regulator of lincomycin BGC, FIG 6 ABCF proteins encoded in biosynthetic gene clusters are putative regulators of antibiotic production in response to antibiotics that share a ribosomal binding site. (a) Scheme of the antibiotic-LmrC-LmbU signaling cascade identified in this study. The production of the LmrC protein is induced by ribosome-bound LS A P antibiotics via a ribosome-mediated attenuation mechanism, and it is coordinated with the LmrB and LmrA resistance proteins, which individually reduce the amount of ribosome-bound antibiotic. LmrC then transduces the antibiotic signal from the ribosome to the transcription of lmbU. The LmbU transcriptional regulator activates the expression of subordinate biosynthetic genes (24). (b) Phylogenetic tree of ABCF proteins from 14 representative streptomycete genomes and ABCF proteins from previously characterized BGCs. ABCF proteins from characterized BGCs are marked with the name of produced antibiotic: symbols in the legend indicate the antibiotic group. Note the correlation between the antibiotic group and the ABCF subfamily. The genomic ABCF protein sequences were taken from previously published data (22). A list of streptomycete genomes and BGCs is available in Table S1. (c) LmrC domain architecture combines features of resistance and regulatory ABCF proteins. The presence of the arm domain, resembling the ABCF translation regulator EttA, indicates the regulatory function of LmrC, while the antibiotic resistance domain (ARD) is shared with other structurally characterized ABCF resistance proteins. The peptidyl tRNA interaction motif (PtIM) in EttA and ARD structural motifs refers to a linker that separates two ATP binding domains. The ARD domain is significantly longer than PtIM, allowing direct interaction with PTC. AdpA lin , activates the entire lincomycin BGC independently of the external lincosamide and thus appears to be the principal regulator of lincomycin biosynthesis (28). What would then be the purpose of the LmrC-LmbU signaling cascade discovered here? This signaling cascade may, in response to the extracellular lincomycin secreted by neighboring cells, induce a premature onset of lincomycin production to ensure its synchronous biosynthesis in a wider population, thereby achieving an ecologically efficient lincomycin concentration similar to that proposed for the biosynthesis of the cell wall inhibiting lantibiotic planosporicin (49) or actinorhodin (50). In addition, we have also shown that the LmrC-LmbU signaling cascade and thus lincomycin production might be activated by functionally similar LS A P antibiotics produced by other organisms. Thus, analogous ABCF signaling cascades (Fig. 6b) could coordinate the production of the same types of antibiotics across different organisms sharing a single niche and so mediate cooperative interspecies interactions (51). In support of this concept, a recent study showed that antibiotic production is more likely to be induced by closely related strains or strains sharing BGCs (5). These observations also imply that the ability of antibiotics to induce their own synthesis is a relatively widespread but mostly undetected phenomenon because antibiotic production in the presence of cognate or similar antibiotics is not usually examined.
The induction of specialized metabolism by antibiotics targeting the 50S subunit of the ribosome has been described previously (6,7,52), but this is the first time the mechanism of antibiotic sensing and signal transduction has been revealed. The detection of antibiotics by a ribosome via a 59UTR attenuator upstream of the ABCF-encoding gene differs fundamentally from known antibiotic signaling cascades, in which antibiotic or biosynthetic intermediates are detected regardless of their mode of action, typically by direct binding to a transcription factor or its cognate receptor (53). In addition, several examples of resistance systems consisting of antibiotic efflux and cognate, TetR-like transcriptional repressors, such as in the biosynthesis of simocyclinone (54), actinorhodin (50), or landomycin A (55), have been described to promote antibiotic production by sensing a final product or intermediate. The regulatory effect in these examples is facilitated by the export of antibiotics, which is required for high production (50, 54), or is mediated by a cognate antibiotic-recognizing repressor that, in addition to the regulation of transporter, also regulates biosynthetic genes (55). In contrast, LmrC appears to directly transduce the antibiotic signal to lmbU transcription while conferring low antibiotic resistance. However, the exact mechanism of LmrC-driven signal transduction remains to be elucidated. Thus, LmrC has a dual function: resistance and regulation, but it is also possible that the low LmrC-mediated resistance is only an indirect consequence of the primary antibiotic signal transduction function and that its biological significance is minor. Notably, LmrC, as well as other ABCF proteins implicated in antibiotic resistance in streptomycetes, shares the antibiotic resistance domain ARD with structurally characterized antibiotic resistance ABCF proteins, VmlR, MsrE, VgaA LC , VgaL, and LsaA from Firmicutes. The ARD interacts with PTC to dislodge the antibiotic from the ribosome (34, 35,56,57) (Fig. 6c), and it is present in the majority of antibiotic-resistant ABCF proteins but not in EttA and other putative regulatory ABCFs (22). In addition to ARD, LmrC also has the arm domain, which is absent in antibiotic-resistance ABCFs but is present in the EttA translation regulator (20). In EttA, the arm domain restricts ribosome dynamics in response to a lack of available ATP (21). However, further research will be needed to determine whether all the ABCF proteins structurally similar to LmrC, i.e., having both the ARD and the arm domain, have regulatory rather than resistance functions. In addition to LmrC, another two ABCF proteins were induced by clindamycin in S. lincolnensis (Fig. 3a), which is a strong indication that ABCF proteins not associated with BGCs for PTC-targeting antibiotics may also have an antibiotic-responsive regulatory function.
The signaling pathway described here, in which the antibiotic signal is sensed and transduced by the dual, resistance, and regulatory ABCF proteins and tuned by two other resistance proteins, points out the need to reconsider the role of antibiotic resistance ABCF proteins as purely protective mechanisms. This discovery also brings together two functionally inconsistent groups of ABCF proteins, antibiotic resistance and regulatory proteins (22,58), which fundamentally changes the view of these translational ATPases. In addition, given the number of small molecules targeting the 50S ribosomal subunit and the number of bacterial ABCFs encoded by soil bacteria from the Terrabacteria group, which includes Firmicutes and Actinobacteria with the highest number of ABCFs per genome, ABCF-mediated signaling could be one of the most important tools of chemical communication in general.

MATERIALS AND METHODS
Bacterial strains and growth conditions. The strains, plasmids, and oligonucleotides used in this study are listed in Table S2. Streptomyces strains were grown at 30°C on solid MS medium (59) Table S2. PCR targeting was applied to the cosmid LK6 (23), which contained the entire lincomycin biosynthetic cluster. After conjugation of mutated cosmids into S. lincolnensis, kanamycin-sensitive (Kan s ) and apramycin-resistant (Apr r ) double-crossover mutants with target genes replaced by the aac(3)IV-oriT cassette were confirmed by PCR amplification. For the construction of S. lincolnensis DAB (BN3021), S. lincolnensis DAC (BN3018), and S. lincolnensis DBC (BN3008) double mutants, the inactivation cassettes 773 in S. lincolnensis DB (BN3002) and S. lincolnensis DC (BN3001) single mutants were replaced by an unmarked in-frame deletion obtained by FLP-mediated excision of the disruption cassette (62). The second gene to be deleted was replaced with cassette 775 according to the same protocol as used for single-knockout strains. Knockout strains were verified by PCR and Southern blot analysis. The scheme of the orientation of inactivation cassettes in all knockout strains is available in Fig. S1.
Construction of vectors for natural, constitutive, or inducible expression. Details on the preparation of vectors and oligonucleotides used are listed in Table S2. The vectors A n (bearing lmrA with its 1,330-bp upstream sequence), Bn (bearing lmrB gene with an 86-bp upstream region), and Cn (bearing lmrC gene and its 1,281-bp upstream region) were used to express resistance genes under its natural promoter. For constitutive expression, lmrA, lmrB, and lmrC were PCR amplified from ligated under the ermEp promoter of pIJ10257 (64), yielding plasmids A C , B C , and C C . For the coexpression of lmrA or lmrB with lmrC in the heterologous host, lmrC with the ermEp promoter was cloned into PtipA expression vector pIJ6902 (65) (construct C c2 ). All the above-mentioned vectors were prepared by restriction enzyme cloning. Constructs C C -mCh, C-U-mCh, U-mCh, U C -mCh, 59C-mCh, and C-mCh were prepared by using the SLICE cloning method (66) (for details, see Table S2). Inducible expression in vectors C i and C EQ12i were achieved by introducing the theophylline-dependent riboswitch (67) via the whole plasmid PCR (details in Table S2). All the constructs described were verified by sequencing.
Site-directed mutagenesis. To introduce mutations, the QuikChange protocol (Agilent) was used. In plasmid C EQ12 , for expression of the lmrC EQ12 mutant, two mutations were introduced into the lmrC coding sequence: glutamate 167 was replaced with a codon for glutamine, and the parallel codon for glutamate 495 was replaced with a codon for glutamine. To test the putative terminator structure, a series of G-to-C and C-to-G point mutations were introduced into the lmrC upstream region in C-mCh vector, yielding plasmids pGBN120, pGBN121, and pGBN124. To test the function of uORFs 1 to 4, their START codons were mutated in C-mCh, yielding plasmids pGBN070 (uORF 1), pGBN106 (uORF 2), pGBN117 (uORF 3), and pGBN119 (uORF 4). All used oligonucleotides are listed in Table S2. All the constructs described were verified by sequencing.
Antibiotic susceptibility tests. The MIC values were determined on MH agar with a serial 2-fold dilution of antibiotics. Frozen spores (see Fig. S7a) or mycelia from 42-h seed culture or 120-h production culture (see Fig. S7b) were diluted in 2 ml of sterile water to optical density OD 450 0.2 to 0.3, and 5 ml was spotted on MH agar with antibiotic and incubated at 30°C for 5 days.
Extraction of lincomycin and celesticetin. A total of 1 ml of supernatant from 42-h seed culture or 160-h production culture (see Fig. S7b) was used for solid-phase extraction as follows: an Oasis HLB 3-ml 60-mg cartridge (hydrophilic-lipophilic balanced sorbent; Waters, USA) was conditioned with 3 ml of methanol and equilibrated with 3 ml of water, and then 1 ml of the supernatant of cultivation broth (for lincomycin extraction pH adjusted to 9.0 with ammonium hydroxide) was loaded. The cartridge was washed with 3 ml of water, and absorbed substances were eluted with 1.5 ml of 80% methanol. The eluent was evaporated to dryness, reconstituted in 150 ml of 50% methanol, and centrifuged at 12,045 Â g for 5 min at room temperature. The extract was then diluted 10Â with methanol-water (1:1 [vol/vol]) and analyzed by liquid chromatography-mass spectrometry (LC-MS), as described below.
LC-MS analysis of lincomycin and celesticetin. LC analyses of the samples depicted in Fig. 1 and in Fig. 3a were performed on an Acquity UPLC system equipped with a 2996 DAD detector and LCT premier XE time-of-flight mass spectrometer (Waters). Five microliters of each sample was loaded onto an Acquity UPLC CSH C 18 LC column (50 mm Â 2.1 mm, inner diameter [ID]; particle size, 1.7 mm; Waters) kept at 40°C and eluted with a two-component mobile phase. For phases A and B, the A solution was 1 mM ammonium formate (pH 9) for lincomycin detection (prepared by titration of formic acid 98 to 100% [Merck, Germany] with ammonium hydroxide 28 to 30% [Sigma-Aldrich, Germany]), the A solution was 0.1% formic acid for celesticetin detection, and the B solution was acetonitrile (LC-MS grade; Biosolve, Netherlands). The analyses were performed with a linear gradient program (min/%B): 0/5, 1.5/ 5, and 12.5/58, followed by a 1.5-min column cleanup (100% B) and 1.5-min equilibration (5% B) at a flow rate of 0.4 ml min 21 . The DAD detector monitored the column effluent in the range 194 to 600 nm; the mass spectrometer operated in the "W" mode with its capillary voltage set at 12,800 V, its cone voltage at 140 V, its desolvation gas temperature at 350°C, an ion source block temperature at 120°C, cone gas flow at 50 liters h 21 , desolvation gas flow at 800 liters h 21 , a scan time at 0.15 s, and an interscan delay at 0.01 s. The data were processed by MassLynx V4.1 (Waters). UV chromatograms monitored at 194 nm were used for lincomycin quantitation based on a five-point linear calibration curve, which was constructed from peak areas corresponding to lincomycin. Calibration solutions were prepared by spiking lincomycin authentic standard at the required concentration into lincomycin-free cultivation broth, extracted and preconcentrated as described above. The quantitation parameters were as follows: concentrations used for the calibration curve were 3.78, 7. 56, 15.125, 31.250, 62.5, and 125 mg liter 21 , the correlation coefficient was r 2 = 0.995, and the limit of quantification was 7.56 mg liter 21 (determined as the lowest point of the calibration curve with precision within 10%). Samples from 42 h of cultivation with lincomycin concentrations below the limit of quantitation were examined by MS detection: extracted ion chromatograms at m/z 407.2 were evaluated for the presence of lincomycin. The 160-h samples for celesticetin production were also examined using MS detection: extracted ion chromatograms at m/z 528.6 were evaluated for the presence of celesticetin.
The LC-MS analyses depicted in Fig. 2b and Fig. S2 were performed on a 6546 LC/Q-TOF (Agilent Technologies, USA) connected to a 1290 Infinity II LC system. One microliter of the sample was loaded on a UPLC CSH C 18 Premier column (100 mm Â 2.1 mm, ID; particle size, 1.7 mm) kept at 30°C. The analytes were eluted at a flow rate of 0.4 ml min 21 with a two-component mobile phase consisting of 1 mM ammonium formate (pH 9) (A) and acetonitrile (B) using a linear gradient program min/%B: 0/5, 1.5/5, 15/65, 15.1/100, 16/100, 16.1/5, and 17.5/5. The mass spectrometer operated in ESI1 mode (jet stream technology) with the following settings: capillary voltage, 3,500 V; nozzle voltage, 200 V; gas temperature, 250°C; drying gas, 8 liters min 21 ; nebulizer, 35 lb/in 2 ; sheath gas temperature, 400°C; sheath gas flow, 12 liters min 21 ; fragmentor, 140 V; and skimmer, 65 V. The ions of m/z 80 to 1,200 were monitored with scan rates of 4 spectra s 21 and 250 ms/spectrum. The identity of the analytes was confirmed by the comparison of retention times with an authentic standard, accurate mass, and collision-induced dissociation fragmentation at a collision energy of 20 eV. The data were processed using Quantitative 10.1 software within a MassHunter workstation (Agilent). Lincomycin quantitation was performed using a standard calibration curve of the lincomycin standard (2-fold serial dilutions from 0.0097 to 40 mg liter 21 ) dissolved in the solid-phase extract of a lincomycin nonproducing Streptomyces lincolnensis lmbD deletion mutant strain. The LLOQ (lower limit of quantification) was determined as the lowest analyte concentration determined with sufficient precision (relative standard deviation of 20%) and accuracy (80 to 120%) using a calibration curve with its lowest point being equal to LLOQ.
Protein digestion for proteomic analysis. Mycelia (0.1 g) of 40-h seed culture inoculated from fresh spores (see Fig. S7b) were lysed in 0.5 ml of 100 mM triethylammonium bicarbonate buffer (pH 8.5) containing 2% sodium deoxycholate, 10 mM Tris(2-carboxyethyl)phosphine, and 40 mM chloroacetamide and boiled at 95°C for 5 min. Protein concentration was determined using a BCA protein assay kit (Thermo), and 20 mg of protein per sample was used for MS sample preparation. Samples were digested with trypsin (at a trypsin/protein ratio of 1/20) at 37°C overnight. After digestion, the samples were acidified with trifluoroacetic acid (TFA) to a final concentration of 1%. Sodium deoxycholate was removed by extraction to ethyl acetate (68), and peptides were desalted on a Michrom C 18 column.

SUPPLEMENTAL MATERIAL
Supplemental material is available online only.