ClpP inhibitors are produced by a widespread family of bacterial gene clusters

The caseinolytic protease (ClpP) is part of a highly conserved proteolytic complex whose disruption can lead to antibacterial activity but for which few specific inhibitors have been discovered. Specialized metabolites produced by bacteria have been shaped by evolution for specific functions, making them a potential source of selective ClpP inhibitors. Here, we describe a target-directed genome mining strategy for discovering ClpP-interacting compounds by searching for biosynthetic gene clusters that contain duplicated copies of ClpP as putative antibiotic resistance genes. We identify a widespread family of ClpP-associated clusters that are known to produce pyrrolizidine alkaloids but whose connection to ClpP has never been made. We show that previously characterized molecules do not affect ClpP function but are shunt metabolites derived from the genuine product of these gene clusters, a reactive covalent ClpP inhibitor. Focusing on one such cryptic gene cluster from Streptomyces cattleya, we identify the relevant inhibitor, which we name clipibicyclene, and show that it potently and selectively inactivates ClpP. Finally, we solve the crystal structure of clipibicyclene-modified Escherichia coli ClpP. Clipibicyclene’s discovery reveals the authentic function of a family of natural products whose specificity for ClpP and abundance in nature illuminate the role of eco-evolutionary forces during bacterial competition. Targeted-directed genome mining identified a widespread family of bacterial ClpP-associated clusters whose active products previously eluded detection. One of these clusters from Streptomyces cattleya produces clipibicyclene that selectively inactivates ClpP and may play a role in bacterial competition.


Target-directed genome mining for ClpP-directed compounds.
To avoid self-intoxication, BGCs contain resistance genes for protection from their encoded antibiotic. When this resistance gene takes the form of an antibiotic-insensitive copy of the targeted protein, searching for BGCs containing that target gene can lead to the identification of new antibiotics. For example, so-called 'target-directed genome mining' was applied in the discovery of thiolactomycin inhibitors of fatty acid synthases FabB /F 16,17 . The ADEP BGC encodes a resistant copy of ClpP 13 , while indiscriminate protease inhibitors such as antipain and chymostatin encode no known resistance mechanism (that is, protease) other than export 14,18 , suggesting that the presence of ClpP in a BGC may indicate a specific inhibitor.
To apply target-directed genome mining to ClpP, we used Actinobacterial ClpPs as queries and searched RefSeq for gene clusters in close genetic proximity ( Fig. 1a; Supplementary Note). We filtered out housekeeping clpPs coincidentally located next to BGCs, landing on a final prioritized list of ten ClpP-associated gene clusters ( Fig. 1a and Supplementary Table 1). Our approach is similar to that of the antibiotic resistant target seeker (ARTS) 17 but while the latter identifies any duplicated essential gene within a set of genomes' BGCs, ours focused on a single target, ClpP, and searches all genomes in RefSeq.
Interestingly, six of our ten final prioritized BGCs contained a bimodular non-ribosomal peptide synthetase (NRPS) predicted to activate proline and serine. We identified 306 BGCs in unique species' genomes from RefSeq with this bimodular NRPS (Supplementary Table 2). These BGCs were missed in our initial analysis due to cluster fragmentation, upload date or lack of ClpP association. As a comparison to these BGC's frequency, streptothricin-family BGCs, which are estimated to be produced by one in ten antibiotic-producing Streptomyces spp. 19 , are present in RefSeq only 50 times. In addition to its abundance, this bimodular NRPS is widespread across genera of the order Actinomycetales (Streptomyces, Saccharothrix, Nocardia and so on) as well as across phyla in important human pathogens, including Pseudomonas aeruginosa, Acinetobacter baumannii and Enterobacter cloacae (Supplementary Table 2). While not all the identified BGCs were associated with ClpPs, many were adjacent to other serine hydrolases such as S8-family peptidases and β-lactamases. These results   Table 3 and an unrelated BGC identified in Amycolatopsis sp. CA-126428 marked X is used to root the tree. c, Schematic of cac identified in S. cattleya DSM 46488 and putative gene functions. AZC, L-azetidine-2-carboxylic acid; PKS, polyketide synthase. d, Activation of cac, as assessed by HPLC chromatograms of n-butanolic extracts from engineered S. coelicolor M1154 strains: (I) the empty shuttle vector (pCGW), (II) the shuttle vector carrying cac plus overexpressing the transcriptional activator Cac15 (pCGW-cac pIJ10257-cac15) or (III) the fully refactored cac cluster (pCGW-cac-LHK). Only (III) results in production of new metabolites, while the lack of metabolites in (II) is consistent with incomplete transcriptional activation of the cluster in this strain (Extended Data Fig. 1b). e, Numbered peaks were identified as azabicyclene family compounds or hydrolysed acyl tails. Fermentations were performed in triplicate and a representative chromatogram is shown.
suggest that this bimodular NRPS BGC family is unusually common, widespread in environmental and clinical bacteria and may have a conserved function of inhibiting Ser hydrolases. We selected 30 exemplar BGCs and built a multilocus phylogeny using CORASON (core analysis of syntenic orthologues to prioritize natural product BGCs) 20 . This analysis revealed three distinct groups among the collection of BGCs (Fig. 1b) characterized by presence of (1) a Baeyer-Villiger flavin-containing monooxygenase (FMO) in a pseudomonal strain (produce the phospholipase inhibitor SB-253514, also called brabantamide; Extended Data Fig. 1a) 21,22 , (2) a condensation domain-containing protein (products not characterized) or (3) an FMO in addition to ClpP. This third clade is known to produce a family of pyrrolizidine alkaloids, the biosynthesis of which has been characterized for compounds such as bohemamines 23 and azabicyclenes (also called azetidomonamide B; Extended Data Fig. 1a) 24,25 . However, the obligate association of these BGCs with ClpP has not been previously explored, which we thought striking given >50 publications on the family over 40 years since their discovery in 1980 26 .
We became interested in a cluster from Streptomyces cattleya DSM 46488 that, in addition to the bimodular NRPS, also contains a type I polyketide synthase ( Fig. 1c and Supplementary Table 4). We named the cluster cac for ClpP-associated cluster. The cluster was transcriptionally silent under several growth conditions tested including following overexpression of the cluster-situated transcriptional activator ( Fig. 1d and Extended Data Fig. 1b). We successfully achieved production using heterologous expression of cac in the superhost Streptomyces coelicolor M1154 from a refactored cluster with non-native promoters (pCGW-cac-LHK; Fig. 1d and Extended Data Fig. 1c).
We purified and solved the structures of major compounds produced by the heterologous host, S. coelicolor  Tables 5-8 and Supplementary Discussion). The primary products identified were azabicyclenes with a hydroxylated decatriene acyl tail (Fig. 1e, compound 5) and related congeners (compounds 4 and 6) or primary amide variants of the respective acyl tails (compounds 1-3). We named these molecules azabicyclenes B (4), C (5) and D (6), following the nomenclature of the related compound azabicyclene (which we hereto refer to as azabicyclene A) produced by the ClpP-associated BGC aze in P. aeruginosa PAO1 (Extended Data Fig. 1a) 24,25 . The biosynthesis of azabicyclenes B-D is predicted to be similar to bohemamines and azabicyclene A 23,25 (Supplementary Discussion and Extended Data Fig. 3).
An elusive product of the cac BGC results in ClpP inhibition. We next tested whether azabicyclenes disrupt the function of ClpP. However, azabicyclene C and D did not robustly inhibit the growth of eukaryotic cells or bacterial species where ClpP is essential, nor impact ClpP activity in vitro (Extended Data Fig. 4a,b). Similarly, purified bohemamines did not affect ClpP (Extended Data Fig. 4c), consistent with no previously reported biological activity 27 . We hypothesized that an undetected product of the BGC might instead impact ClpP function. Therefore, to comprehensively assay metabolites produced by cac, we probed ClpP function directly in S. coelicolor M1154 pCGW-cac-LHK. In S. coelicolor, as in many Streptomyces spp., there are five copies of clpP in three operonic units: clpP1clpP2, clpP3clpP4 and clpP5. Each operon encodes subunits capable of forming a functional hetero-tetradecameric complex except for ClpP5, whose catalytic triad is irregularly spaced and proteolytic activity unknown 28 . ClpP1P2 complexes are susceptible to ADEP activation while ClpP3P4 is resistant (Fig. 2a) 29 . Therefore, if ClpP1P2 is functional, non-specific degradation induced by ADEP results in cell death, while if ClpP1P2 is inhibited, ClpP3P4 is expressed and provides ADEP resistance. S. coelicolor M1154 pCGW-cac-LHK was more resistant to ADEP than the empty vector control, S. coelicolor M1154 pCGW, indicating ClpP1P2 inhibition in the heterologous producer (Fig. 2b). Deleting the cluster-associated ClpPs homologous to clpP1 and clpP2, cac16 and cac17 respectively, provided further ADEP resistance. This finding supports that Cac16/17 is susceptible to ADEP but resistant to inhibition by cac BGC products (Fig. 2a).
To detect secreted metabolites from the heterologous producer that inhibit ClpP1P2, we made use of the transcriptional regulation of ClpP homologues in S. coelicolor. Under normal conditions, ClpP1P2 is the primary housekeeping complex and targets PopR, the activator of clpP3clpP4, for degradation ( Fig. 2c) 28,30,31 . Therefore, the clpP3clpP4 operon is only activated when ClpP1P2 is disrupted 28 . Correspondingly, we observed clpP3clpP4 activation in S. coelicolor M1154 pCGW-cac-LHK (Extended Data Fig. 5a). Therefore, we constructed a reporter strain for ClpP1P2 inhibitors where the reporter gene gusA 32 is driven by the clpP3 promoter in wild-type S. coelicolor. Costreaking this strain next to S. coelicolor M1154 pCGW-cac-LHK resulted in P clpP3 activation, which was abolished by deletion of a biosynthetic gene ( Fig. 2e and Extended Data Fig. 5b,c). P clpP3 was not stimulated by purified azabicyclenes C and D or bohemamines (Extended Data Fig. 5d) but was activated by costreaking the bohemamine producer itself (Extended Data Fig. 5e). Therefore, the production of secreted ClpP inhibitors is a widespread trait of this family of BGCs and our P clpP3 :gusA reporter strain may be a useful tool for identifying ClpP inhibitors in the future.
Next, we tested whether our elusive ClpP1P2 inhibitor could kill actinobacterial species where ClpP is essential. Agar plugs inoculated with S. coelicolor M1154 pCGW-cac-LHK inhibited growth of Mycobacterium smegmatis but only minimal zones of inhibition were observed against wild-type S. coelicolor (Fig. 2e). We hypothesized that these small zones resulted from intrinsic resistance provided by ClpP3P4, as has been observed for other β-lactone ClpP inhibitors 4 . Indeed, S. coelicolor ΔclpP3P4 was more susceptible than wild type to growth inhibition (Fig. 2e). In a collection of six other Streptomyces spp., those that naturally lacked clpP3clpP4 were overall more susceptible (Extended Data Fig. 5f), suggesting that it is a general resistance mechanism in the environment.
ClpP inactivation guides purification of clipibicyclene. Given our strong evidence for production of a ClpP inhibitor by cac family BGCs, we wondered why we and others were previously unable to detect this bioactivity or the responsible metabolite. The ClpP1P2 inhibitory activity of spent media from S. coelicolor M1154 pCGW-cac-LHK decreased throughout a 5-day fermentation and was lost during incubation overnight at room temperature or n-butanol extract generation (Extended Data Fig. 5g). In contrast, azabicyclenes accumulate in fermentation broth over time, suggesting that they are shunt metabolites. We hypothesized that the active compound may possess a reactive warhead that inhibits ClpP by covalent modification of the active site Ser and that its resulting instability made it previously difficult to detect.
To test for covalent modification of ClpP, we first reconstituted the in vitro activity of recombinant S. cattleya ClpP1 and ClpP2 hetero-tetradecamers (ClpP1P2 scatt ). Peptidase activity required both subunits to be present and was stimulated by ADEP (Fig. 3a), similar to M. smegmatis ClpP1P2 (ref. 33 ). We characterized how ClpP1P2 scatt is N-terminally processed, the role of agonist peptides, substrate selection and asymmetric interaction with AAA+ ATPases (Supplementary Discussion and Extended Data Fig. 6).
Next, we incubated recombinant ClpP in fresh spent media from S. coelicolor M1154 pCGW-cac-LHK containing the active metabolite. After recovery from spent media, the activity of both ClpP1P2 scatt and Escherichia coli ClpP (ClpP ec ) was abolished (Fig. 3b) and analysis by intact protein liquid chromatography-mass spectrometry (LC-MS) revealed a mass increase of ClpP1 scatt and ClpP ec of ~328 Da (Fig. 3c). Peptide mapping of ClpP1 scatt identified the same 328 Da modification on the active site serine and fragments observed from tandem mass spectrometry (MS/MS) analysis were consistent with the acyl tail of azabicyclenes ( Fig. 3d and Extended Data Fig. 2d).
To identify the active metabolite, we analysed spent media by LC-MS and detected a peak with an m/z of 329.14 Da, which corresponds to the mass shift observed from the modification of ClpP. This peak was associated with a parent ion with an m/z = 347.15, suggesting m/z = 329.14 Da is a dehydration product (Fig. 3e). This species was removed from spent media after incubation with recombinant enzyme, confirming that it was the active metabolite (Extended Data Fig. 7a). We subsequently isolated the active compound, which we named clipibicyclene (Fig. 3f, compound 7) and found it was relatively stable once removed from the fermentation broth (Extended Data Fig. 7b).
Complete structural elucidation of clipibicyclene revealed a bicyclocarbamate warhead attached to an acyl tail matching azabicyclene C (Fig. 3f, Supplementary Table 9, Supplementary Fig. 5 and Supplementary Discussion). A similar bicyclocarbamate was previously identified from the ClpP-associated BGC in P. aeruginosa PAO1 (Extended Data Fig. 1a) but was not recognized as a ClpP inactivator 24 . To account for the difference between the mass of clipibicyclene and the observed mass shift from intact protein LC-MS, we propose a mechanism whereby clipibicyclene's reactive carbamate moiety undergoes nucleophilic attack by the active site Ser of ClpP and results in elimination of the hydroxyl group attached to the azetidine ring (Fig. 3g).
Pure clipibicyclene is an efficient ClpP inactivator, as residual enzyme activity is abolished at an inactivator to enzyme ratio ([I]/ [E]) of 1.2 for ClpP ec (Fig. 4a). Remarkably, unlike many other promiscuous protease inactivators, four other proteases tested were not inhibited by clipibicyclene, indicating specificity to ClpP (Fig. 4b). We also assessed specificity on whole-cell S. coelicolor lysates by profiling Ser hydrolases using the fluorescent probe, TAMRA-fluorophosphonate (TAMRA-FP). In contrast to the promiscuous inactivators phenylmethylsulfonyl fluoride (PMSF) and ebelactone A, no off-target activity was detected for clipibicyclene up to 100 μM (Fig. 4c).
Finally, we assessed the bioactivity of pure clipibicyclene. Despite activity in agar plug assays (Fig. 2e), pure compound did not inhibit growth up to 512 μg ml -1 , possibly owing to instability over the >24 h incubations required for these assays (Extended Data Fig. 7b). We envision that in a natural environment, the continual production Cac16/17 A c ti v a te s D e g r a d e s Term T7

Fig. 2 | biological activity of an elusive ClpP inhibitor. a,
Summary of ClpP isoforms present in S. coelicolor M1154 pCGW-cac-LHK and their susceptibility to ADEP activation or cac metabolite inhibition. b, Kirby-Bauer assays with ADEP show the sensitivity of S. coelicolor M1154 strains harbouring the constructs indicated. Complementation of cac16-17 is provided by the plasmid pIJ-ncac16-17. Each disk contains 50 μg of ADEP. c, Schematic of ClpP regulation by the ClgR regulon in S. coelicolor. ClgR is a transcriptional activator that regulates the clpP1clpP2 operon as well as its own expression 30,31 . Subsequently, ClpP1/P2/X degrades ClgR and the transcriptional activator for the clpP3clpP4 operon, PopR. A similar negative feedback loop occurs between ClpP3P4 and PopR 28 . The promoter region of cac16-17 contains a ClgR binding motif 31 and is also likely to be regulated by ClgR. Orange lines indicate protein degradation and black arrows indicate transcriptional activation. d, Reporter strain activation of P clpP3 . The schematic shows the design of the plasmid pIJGUS-pClpP3, which expresses GusA, a β-glucuronidase that hydrolyses the chromogenic substrate X-gluc resulting in the production of a blue pigment. Costreaking S. coelicolor M1154 strains harbouring constructs for heterologous production (streaked horizontally) next to indicator strains (streaked vertically) reveals P clpP3 activation as indicated by the blue pigment. Deletion of a biosynthetic gene, cac8, but not cac16-17, abolishes P clpP3 induction. Indicator strains also contain empty vector (pSET152) or that for expressing cac16-17 (pSET152-cac16-17). e, Kirby-Bauer assays for growth inhibition make use of agar plugs inoculated with S. coelicolor strains as indicated on the left. Strains that are being tested for susceptibility are labelled along the top. Complementation of clpP3clpP4 expression is provided by the plasmid pIJ-clpP3clpP4. WT, wild type. Experiments in b, d and e were performed on at least two independent occasions with similar results.
of clipibicyclene, as reflected in agar plug assays, could inhibit the growth of neighbouring species to provide a competitive advantage.

Cluster-associated ClpPs are resistant to clipibicyclene.
A fundamental postulate of ClpP-directed genome mining is that cluster-associated ClpPs, in this case Cac16 and Cac17, provide resistance to the produced antibiotic. Consistently, expression of cac16-17 in S. coelicolor provided resistance to growth inhibition by clipibicyclene (Fig. 5a). It also protected cells from P clpP3 activation ( Fig. 2d), demonstrating that Cac16/Cac17 performs an analogous function to ClpP1P2 (Fig. 2b).
To investigate clipibicyclene resistance in vitro, we reconstituted the activity of recombinant Cac16 and Cac17 and characterized their activity and processing similar to ClpP1P2 scatt (Fig. 3a, Extended Data Fig. 6 and Supplementary Discussion). We assessed the susceptibility of different ClpP isoforms to covalent modification by intact protein LC-MS (Fig. 5b). ClpP1 scatt was modified to a greater extent than ClpP2 scatt and ClpP2 scatt was not modified at   all when co-incubated with ClpP1 scatt . Cac16 was almost wholly resistant to modification by 100 μM clipibicyclene, while Cac17 was partially modified. Next, we titrated clipibicyclene against either ClpP1P2 scatt or Cac16/17 and measured peptidase activity (Fig. 5c) . 5c). Together with cell-based assays, these results show that Cac16/17 provide resistance to clipibicyclene through reduced susceptibility to covalent inactivation. We wondered whether both Cac16 and Cac17 are required for resistance to clipibicyclene. To this end, we tested growth inhibition of S. coelicolor ΔclpP3P4 expressing various combinations of Cac16, Cac17 and catalytically dead mutants (Cac16 S108 , Cac17 S129A,S130A ). Cac16 alone was necessary and sufficient to provide resistance to clipibicyclene and required that the enzyme be catalytically active (Fig. 5a). In contrast, Cac17 was insufficient to provide resistance. To explain this discrepancy, we investigated the minimal ClpP subunits required to support the growth by engineering S. coelicolor for expression of specific ClpP isoforms in a background lacking other housekeeping homologues (Fig. 5d). Both ClpP1 S99A /ClpP2 and ClpP1/ClpP2 S132A supported growth (Fig. 5e), showing that only a single catalytically active ClpP isoform is required. Co-expression of Cac16/17 supported growth, as did Cac16/Cac17 S129A,S130A but Cac16 S108 /Cac17 did not (Fig. 5e). Therefore, Cac17 lacks the required catalytic activity or biological function for it to support growth. Combined with in vitro data showing partial modification of Cac17 by clipibicyclene, these observations explain Cac17's inability to provide resistance.
Finally, we wondered whether functional tetradecamers were formed strictly between ClpP1/ClpP2 and Cac16/Cac17 or if 'mixed complexes' of ClpP1/Cac17 and Cac16/ClpP2 were also relevant for resistance and growth. Mixed complexes were active in in vitro peptidase assays (Fig. 5f) and in vivo could support growth of S. coelicolor (Fig. 5e). Only Cac16/ClpP2 provided resistance, consistent with our previous results (Fig. 5a). Therefore, ClpP subunits can mix and match in the cell and these mixed complexes are physiologically relevant.
Crystal structure of ClpP ec in complex with clipibicyclene. We determined an X-ray crystal structure of ClpP ec in complex with clipibicyclene to 2.95 Å resolution with molecular replacement using the apo form of ClpP ec as a search model 34 (PDB 1TYF; Supplementary  Table 10). The ClpP ec inactivator complex crystallized with two complete tetradecamers in the asymmetric unit with all catalytic serine residues (S97) modified (Extended Data Fig. 8a) 34 . The N-terminal residues (7-17) that form each heptameric ring's axial pore were disordered owing to high conformational flexibility 35  be fully modelled (Extended Data Fig. 8b). Each ClpP ec monomer  did not substantially differ from the apo form as judged by superposition of the tetradecamer (root mean square deviation = 0.38 Å over 16,148 atoms; Extended Data Fig. 8c).
The structures of the modified Ser97 residues were built and restrained under clipibicyclene's proposed reaction mechanism (Fig. 3g) and fit well within the electron density map (Fig. 6a). In the active site context, the clipibicyclene adduct points clockwise Cac16 Cac17 Cac16/Cac17 None Cac16/Cac17 S129A,S130A Cac16 S108A /Cac17  The operator sequence cmtO is bound by the repressor CymR, providing strict repression in the absence of cumate. P A26 and P A51 are strong, constitutive synthetic promoters 44 . e, The ability of various ClpP homologues to support growth is tested by expression in S. coelicolor ΔclpP3clpP4 P A26-cmtO :clpP2clpP2 in the presence/absence of cumate. In the absence of cumate, the defined subunits expressed in trans, as indicated, are the only ClpP homologues present in the cell. Since heterocomplexes are required for tetradecamer formation, individual subunits are tested by generating inactive variants of each isoform's partner. f, Mixed complexes of Cac16/ClpP2 and ClpP1/Cac17 form active heterocomplexes in vitro. Complexes containing Cac16 or Cac17 may be less active due to suboptimal reaction conditions. Assays were performed with ADEP using the fluorogenic peptide substrate S-LLVY-AMC and 2 μM of each subunit. For c and f, the mean of technical replicates (n = 2) with error bars representing standard deviation (not visible in f at this scale) is shown. All experiments were performed on at least two independent occasions with similar results.
relative to the interior circumference of the heptameric ring of ClpP ec (Fig. 6b). The azetine ring is positioned between sheets β4 and β9 in the hydrophobic S1 subsite and forms van der Waals contacts with I70, M98, P124 and L125 (Fig. 6c,d). The carbonyl oxygen of the carbamate linkage forms a critical hydrogen bond with the backbone amide of Gly68, which usually serves as an H-bond donor in the oxyanion hole (Fig. 6d). On the basis of this model, irreversible inhibition can be partly explained by both the inherent stability of the carbamoyl moiety and/or steric exclusion of water from the catalytic base (His122).
Clipibicyclene's linear aliphatic chain, which showed a high degree of flexibility, could not be modelled in many cases but three complete adduct complexes could be placed (chains A, O and W), revealing different conformations (Extended Data Fig. 9). In the adduct from chain A, the linear portion extends toward the S′ subsites over the carbonyl carbon across the β4 strand and partially protrudes into the solvent. Only one potential H-bond is formed between the side chain of Q134 and the imide moiety and a van der Waals contact is probably made with P66 and the aliphatic tail. Except for a few outliers, these interactions were observed in the majority of chains. This limited interaction network along the tail probably explains its disorder, implying that the flexible lipid tail is not important for ClpP specificity. Future structure-activity relationship studies with clipibicyclene analogues will be required to determine the motifs important for specificity to ClpP.

Discussion
By applying target-directed genome mining we associate a widespread family of BGCs with the highly conserved and often essential protease ClpP. Despite being a relatively well-known BGC family for their production of pyrrolizidine alkaloids, the function and biologically relevant molecules produced by this group of BGCs was elusive. Here, we characterize clipibicyclene, a representative ClpP inhibitor, as the active metabolite produced by cac.
The abundance and widespread nature of this family of bimodular NRPS BGCs, from Streptomyces to Pseudomonas, is unusual compared to other BGC families and suggests that the metabolites produced are important to bacterial physiology. Understanding their function as ClpP inhibitors illuminates an evolutionary arms race, whereby production of these compounds provides a competitive advantage and so spread to diverse phyla and offers a model for studying why some BGC families are more successful than others. Of the proteases tested, clipibicyclene is potent and specific for ClpP, a rare trait for protease inhibitors. This quality sets clipibicyclene apart from promiscuous natural product inhibitors and implies that clipibicyclene's scaffold has been shaped by natural selection to specifically inhibit ClpP.
The presence of these BGCs in human pathogens/commensals, including P. aeruginosa, K. pneumoniae and E. cloacae, is notable and offers a model to understand how bacteria compete in these contexts. Even though ClpP inhibitors are only lethal to bacteria where ClpP is essential, such as Actinobacteria, it is interesting that proteobacterial strains in which ClpP is dispensable still encode BGCs with associated ClpPs. Therefore, even though ClpP activators are preferrable in the sense that that may kill all bacteria, inhibition may still markedly perturb bacterial fitness. Further investigation is warranted, as production of metabolites in P. aeruginosa PAO1 from the ClpP-associated aze BGC causes reduced growth and virulence in a Galleria mellonella model 24 , reminiscent of the reduced virulence of Staphylococcus aureus ClpP mutants 7 . Future studies are also required to address several aspects of clipibicyclene's biosynthesis, including the formation of clipibicyclene's bicyclocarbamate warhead (Supplemental Discussion).
Numerous antibiotics act on and remodel the transcriptional and translational landscape of the cell and, by doing so, play essential roles during infection, adaptation and survival [36][37][38] . Post-translational regulation is probably no less critical in coordinating cellular activities. Clipibicyclene's discovery is a step towards   understanding how small molecules targeting the proteome mediate bacterial physiology and competition.
ADEP was purified from S. hawaiiensis NRRL15010 as previously described with minor modifications 13  MICs were assessed using standard microbroth dilution protocols. Streptomyces spp. were assayed in TSB, M. smegmatis in 7H9 media and all other strains in Mueller-Hinton Broth (BD Biosciences).
Cloning cac using TAR and refactoring pCGW-cac. Plasmid pCGW is a capture construct derived from pCAP03 (ref. 16 ) and modified to use the 'oriV-ori2-repE-sopABC' single-copy origin of replication from pBAC-lacZ as previously described 40 . Plasmid pCGW was maintained in E. coli EPI300, which controls the expression of trfA required for high copy amplification from oriV with an arabinose-inducible promoter. When necessary for miniprep and plasmid mapping, high copy number was induced using 1 mM arabinose.
The boundaries of cac were defined by comparison with homologous clusters in Streptomyces pini PL19 NRRL B-24728, Streptomyces barkulensis RC1831 and Kitasatospora phosalacinea NRRL B-16228. The 50 base pair (bp) homology arms flanking cac (orf-3 to cac18) were concatenated with an MssI site in between and 18 bp overlaps with the pCGW backbone were added to either end. The gBlock (Integrated DNA Technologies; IDT) was cloned into pCGW linearized with NdeI/XhoI by Gibson assembly (Supplementary Table 12). Before TAR cloning, the capture vector was digested with MssI to release the hooks. TAR cloning was performed using high-molecular-weight genomic DNA from S. cattleya DSM 46488, predigested with XhoI and NdeI, by spheroplast transformation into Saccharomyces cerevisiae VL648N 41 . DNA from positive yeast colonies was transformed and maintained in E. coli EPI300 with 1 mM arabinose, where required, for copy number induction.
Refactoring pCGW-cac. Plasmid pCGW-cac was refactored using a combination of yeast recombination and λ-Red recombineering in E. coli. All primers, gblocks, strains and plasmids are listed in Supplementary Tables 11-13. First, Leu2 and His3 auxotrophic selection cassettes were constructed to contain orthogonal promoters and terminators using pSASS series plasmids, a gift from M. Tyers (University of Montreal). The leu2 open reading frame was PCR amplified from pRS316 and inserted into pSASS5 and his3 was PCR amplified from pYAC10 (M. Tyers) and inserted into pSASS4, using Gibson assembly. Selection cassettes containing the marker along with yeast promoter and terminator sequences were subsequently PCR amplified from pSASS5-leu2 and pSASS4-his3. For replacement of the cac14 promoter, XNR_1700p was PCR amplified from Streptomyces albus genomic DNA 42 and stitched to the leu2 selection cassette with 500 bp homology arms to cac on either side through overlap extension PCR. The resulting refactoring cassette was cotransformed with pCGW-cac into S. cerevisiae SASy31/SASy35 by standard lithium acetate/single-stranded carrier DNA/PEG-mediated transformation 43 . Recombinants were selected on SD-trp-leu plates, screened by colony PCR and the resulting construct, pCGW-cac-L, was recovered and confirmed in E. coli EPI300, similar to initial TAR cloning. Replacement of cac8 and cac9 promoters were achieved similarly using the His3 selection marker sandwiched by synthetic Streptomyces promoters, A26 and A35, incorporated into primer sequences 44 . This time, S. cerevisiae SASy31 transformants were selected using SD-trp-leu-his plates, and the resulting construct, pCGW-cac-LH, was moved to E. coli EPI300.
Since few selection markers were left for use in S. cerevisiae, we chose to use E. coli λ-Red recombineering to replace the cac5 promoter. The promoter kasOp* and T7 terminator were designed adjacent to the apramycin resistance gene aac(3)IV flanked by PmeI/HpaI restriction sites to remove aac(3)IV after recombination. Thirty-nine bp homology arms were added to each end to direct recombination. The recombineering cassette was synthesized as a gBlock by IDT and recombineering was carried out using standard protocols 45 . Finally, aac (3) IV was removed by digestion with HpaI and PmeI followed by intramolecular blunt-end ligation with T4 DNA ligase, generating pCGW-cac-LHK.
RT-PCR measurement of cac and clpP regulon expression. Strains for analysis were inoculated 1:100 from a saturated TSBY seed culture into 60 ml of Bennett's media in 250 ml baffled flasks and pellets were taken at 24 h. Cells were lysed by bead beating mycelium with 4 mm glass beads in 5 ml of TRIzol reagent (Invitrogen) and RNA was extracted using the manufacturer's recommendations. RNA from the resulting aqueous phase was purified using PureLink RNA Mini Kit (Invitrogen). Maxima H Minus First Strand complementary DNA synthesis kit with dsDNase (Thermo Scientific) was used for cDNA synthesis and PowerUp SYBR Green master mix (Applied Biosystems) was used for PCR with reverse transcription (RT-PCR) quantification on a BioRad CFX96 real-time system using primers listed in Supplementary Table 11. Fold-change expression was calculated using the ΔCt method normalized to hrdB expression. Statistical analysis of clpP3 gene expression was performed using GraphPad Prism v.6. Multiple comparisons to pCGW were made using a two-sided Kruskal-Willis analysis with Dunn's test for multiple comparisons (n = 3).
Purification and structural elucidation of clipibicyclene. S. coelicolor pCGW-cac-LHK culture was grown in the same way as for azabicyclene purification, except that the fermentation was harvested after 24 h. After the growth of a 2 l culture, mycelium was removed and the supernatant was extracted twice by a total volume of 4 l of dichloromethane. The organic phases were combined and concentrated gently in vacuo to obtain a yellow oil. The crude extract was resuspended in 200 μl of DMSO and purified on a Waters XSelect CSH preparative C18 HPLC column (10 × 100 mm 2 , 5 µm, 130 Å, 3 ml min -1 , Buffer A water + 0.1% formic acid, Buffer B acetonitrile + 0.1% formic acid)) using the following method: 0-2 min 25% B, 2-19 min 25-100% B, 19-20 min 100% B, 20-21 min 100-25% B. This afforded 2.0 mg of clipibicyclene, which was subject to LC-MS and NMR structural elucidation as described for azabicyclenes (Supplementary Discussion; Supplementary Fig. 5 and Supplementary Table 9). Chemical shifts are reported in ppm referenced to d 6 -DMSO (δ H 2.50 ppm and δ C 39.5 ppm).
Kirby-Bauer assays. Indicator Streptomyces strains were grown for 2-3 d in TSBY media with antibiotic selection as necessary until saturated. Cultures were diluted to optical density OD 600 = 0.1 and streaked on Bennett's agar using a sterile cotton swab. For assays involving ADEP, 50 μg of compound in DMSO was placed on a sterile cellulose disk. For assays involving agar plugs, S. coelicolor pCGW-cac-LHK or S. coelicolor pCGW was inoculated on Bennett's agar from a saturated TSBY starter culture and grown for 24 h before removal of an agar plug that was placed on a plate inoculated with the indicator strain.
GusA indicator strain construction and assays. A cassette consisting of the T7 terminator placed upstream of P clpP3 :gusA was constructed in the pIJ10257 backbone. Using the primers listed in Supplementary Table 11, parts were PCR amplified as follows: clpP3 promoter region lacking its ribosome-binding site from S. coelicolor gDNA, gusA including its ribosome-binding site from pGUS 32 and T7 terminator from pDR3K 46 . PCR products were cloned into pIJ10257 digested with EcoRV using Gibson assembly to generate the construct pIJGUS-pClpP3. The construct was moved into S. coelicolor M1154 in addition to pSET152 (empty vector) or pSET152-cac16-17, cloned between XbaI and EcoRI restriction sites using primers listed in Supplementary Table 11.
Expression and purification of ClpPs. Untagged E. coli ClpP was expressed from E. coli BL21(DE3) pET9a-EcClpP, a kind gift from W. Houry (University of Toronto) and purified as previously described with minor modifications 47 . S. cattleya ClpP1 (SCATT_17350), ClpP2 (SCATT_17340), Cac16 (SCATT_32700) and Cac17 (SCATT_32710) were codon-optimized for expression in E. coli and synthesized as gBlocks by IDT (Supplementary Table 12). Synthesized DNA was cloned into pET-28 digested with NcoI and NotI using Gibson assembly to be expressed with a C-terminal 6xHis tag. Constructs were transformed into E. coli BL21(DE3) 1146D (ΔclpP::cm r ). For protein expression, overnight cultures were inoculated into 1 l of LB with appropriate antibiotics in a 3 l flask, grown at 37 °C until cultures reached OD 600 = 0.6-0.8, then induced with 0.5 mM IPTG. Cultures were grown for 18 h at 17 °C before harvesting cell pellets by centrifugation.
Where indicated, clipibicyclene, azabicyclene or bohemamines were pre-incubated with enzyme for 10 min before substrate addition. The 100 μl reactions were initiated with the addition of substrate, carried out at room temperature and tracked by fluorescence excitation/emission 360 nm/460 nm (AMC) or 485 nm/525 nm (rhodamine 110). Rates were calculated using the slope over the first 15 min of the reaction. The partition ratio (k cat /k inact ) for each ClpP orthologue was determined by clipibicyclene titration where [I]/[E] was varied between 0.1 and 2.8, keeping the enzyme concentration fixed. The enzyme concentrations used were: ClpP ec (2 μM), ClpP1P2 scatt (4 μM or 2 μM) and Cac16/17 (8 μM).
Serine hydrolase profiling. Serine hydrolase profiling was performed with S. ceolicolor M1154. Starter cultures were grown in TSBY for 24 h at 30 °C and were diluted 1/100 in fresh TSBY (50 ml) and incubated for a further 24 h. Cells were harvested by centrifugation (5,000g, 15 min) and were washed twice in lysis buffer (50 mM HEPES, 100 mM NaCl, pH 7.5). The washed cells were then resuspended in buffer supplemented with benzonase (10 units), mutanolysin (20 units) and phosphatase inhibitors. The cells were subsequently lysed by sonication on ice using 15 s on/ off pulses for a total operating time of 5 min. Unbroken cells were sedimented by centrifugation (18,000g, 15 min) and the total protein was quantified with BCA protein assay and the lysate was adjusted to 3 mg ml -1 . To 100 μl aliquots of lysate, PMSF (100 μM), ebelactone A (100 μM), clipibicyclene (100, 10 and 1 μM) or DMSO were added and incubated for 15 min at room temperature. Serine hydrolases were subsequently labelled with TAMRA-FP serine hydrolase probe and incubated for 30 min at room temperature. The labelling reactions were quenched with Laemmli sample buffer and heated to 95 °C for 5 min. Labelled proteins were analysed by SDS-PAGE analysis and detected with fluorescent gel scanning. Total protein was detected with Coomassie brilliant blue.
Thermal shift assays for ADEP binding. Melt curves were performed using the following conditions: 5 μM ClpP subunit (monomer), 5× SYPRO orange dye (5,000× stock, ThermoFisher), 100 μM ADEP (or DMSO), in ClpP reaction buffer. A total 50 μl per well were assayed on a BioRad CFX96 real-time system, ramping from 35 to 95 °C in increments of 0.5 °C for 10 s, reading SYPRO fluorescence after every increment. Melting temperatures were calculated using CFX Maestro software.
In general, 1 μl of ~30-40 μM protein was injected for analysis. Spectra were deconvoluted and figures were generated using UniDec software 48 . To assess cross-processing, subunits were mixed in a 1:1 molar ratio and incubated overnight at room temperature. Processing sites were predicted using ExPASy FindPept.
To label ClpPs in spent media, spent media was harvested after just 24 h, diverging from our standard protocol of fermentation for 5-14 d. The fermentation broth was first filtered through an Amicon Ultra centrifugal filter with 5 K cut-off to remove secreted proteins present in the media and neutralized by adding 50 mM Tris-HCl pH 7.5. Next, 5 μg ml -1 of ClpP, either total (ClpP ec ) or each subunit (ClpP1P2 scatt ), was added and incubated for 1.5 h at room temperature. ClpP protein was recovered and buffer exchanged using an Amicon Ultra centrifugal filter with 30 K cut-off before LC-MS analysis or peptidase assays. To label ClpP with pure clipibicyclene, 20 μM of each ClpP subunit was first incubated in ClpP reaction buffer for 3 h at room temperature to allow processing to occur before the addition of 100 μM clipibicyclene and LC-MS analysis.
Peptide mapping LC-MS/MS. ClpP1 (50 μg) was incubated with spent media filtrate of S. coelicolor pCGW or S. coelicolor pCGW-cac-LHK cultures (10 ml) for 1.5 h at 25 °C and concentrated to 1 mg ml -1 . Protein was precipitated and washed with cold acetone and the protein was dissolved in 50 mM ammonium bicarbonate buffer pH 8 (20 μl). Sequencing-grade trypsin was added (1:25 w/w ratio with ClpP) and the reaction was incubated at 37 °C for 16 h. The resultant peptides were analysed by LC-MS/MS with an Agilent G6550A Q-TOF in positive-ion mode. Data analysis was performed using Mmass (www.mmass.org).
CRISPR-editing and complementation of S. coelicolor ClpPs. CRISPR-editing constructs containing single guide RNA targeting sequences were cloned using the pCRISPomyces-2 backbone using primers listed in Supplementary Table 11 via golden-gate assembly 49 . S. coelicolor clpP3clpP4 homology arms were designed to introduce a scarless in-frame deletion and cloned using Gibson assembly. For inducible control of clpP1clpP2, a cassette consisting of the cumate-responsive repressor cymR, the strong synthetic promoter A26 and the cymR operator sequence cmtO was designed and synthesized as a gBlock (IDT; Supplementary  Table 12) 44,50 . This cassette was PCR amplified using primers with suitable homology arms for Gibson assembly. Construction of the final pCRISPomyces-2 construct was performed in two Gibson assembly steps: first homology arms for clpP1clpP2 were introduced flanking a HindIII site and, second, the promoter cassette was introduced into the HindIII site.
CRISPR-editing was performed as previously described 51 . Briefly, pCRISPomyces constructs were conjugated to S. coelicolor M1154; exconjugants were restreaked and screened using PCR for successful editing; pCRISPomyces-2 was cured by growth at 37 °C and exconjugants were verified again by PCR. Conjugation and maintenance of the clpP1clpP2 edited strain were performed on media containing 100 μM cumate.
Complementation plasmids pIJ-ncac16-17, pIJ-nclpP1P2, pIJ-nclpP3P4 and pIJ-clpP1-cac17 were cloned by PCR amplifying inserts from S. coelicolor (clpP1clpP2 and clpP3clpP4) or S. cattleya (cac16-17) gDNA using primers listed in Supplementary Table 11 before Gibson assembly into pIJ10257 digested with KpnI and HindIII. For pIJ-cac16-clpP2, the plasmid backbone including cac16 was PCR amplified from pIJ-ncac16-17 and clpP2 was PCR amplified from gDNA for Gibson assembly. All remaining constructs, including pIJ-ncac16, pIJ-ncac17 and catalytically inactive variants (pIJ-ncac16d17, pIJ-cac16-17d, pIJ-clpP1dP2, pIJ-clpP1P2d) were cloned using site-directed mutagenesis by PCR amplifying the full constructs followed by DpnI digestion and ligation. We observed that native promoters provided superior expression than the commonly used constitutive Streptomyces promoter ermEp*, so they were retained in these constructs. Since Cac17 contains two adjacent serine residues in its active site, both were converted to alanine to ensure that the enzyme was catalytically inactive.
Crystallization and structure determination of ClpP ec :clipibicyclene. Preparation of ClpP ec :clipibicyclene for crystallization involved incubating purified enzyme (2 mg ml -1 ) with 0.5 mM clipibicyclene (fivefold molar excess) in ClpP reaction buffer for 1 h at room temperature. Intact protein electrospray ionization mass spectrometry (ESI-MS) verified complete covalent modification. Before crystallization, the covalent complex was buffer exchanged into 10 mM HEPES pH 7.5, 100 mM NaCl using a BioGel P6-DG column and was concentrated to 12 mg ml -1 . Crystals of ClpP ec :clipibicyclene were grown at a 2:1 precipitant to protein ratio using 0.1 M sodium acetate pH 5.6, 30% (v/v) MPD and were flash cooled in a N 2 cryo stream.
Data collection was conducted at the Canadian Light Source CMCF-BM (08IB1), Saskatoon, Canada. The X-ray data were processed using autoPROC, XDS and CCP4 (refs. 52-54 respectively). The structure of ClpP ec :clipibicyclene was determined by molecular replacement with Phenix 55 using apo ClpP ec (PDB ID: ITYF) as the search model. Model building and refinement were carried out using Coot 56 and Phenix 55 with TLS groups determined automatically using the TLSMD webserver 57 . Ramachandran statistics were calculated in Phenix using MolProbity, which gave 97.4% total favoured assignments and 0.21% outliers. To ensure that the stereochemistry of the clipibicyclene adduct remained well-restrained during refinement, complete amino acid restraints were generated for a modified serine residue using the GradeWebServer (http://grade.globalphasing.org). Data statistics are listed in Supplementary Table 10. The coordinates and structure factors have been deposited (PDB code 7MK5) in the Protein Data bank, Research Collaboratory for Structural Bioinformatics, Rutgers University, United States. Molecular graphics and analysis were performed using Pymol.
Reporting Summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
All data generated or analysed during this study are included in this published article and its Supplementary Information. The S. cattleya DSM 46488 whole genome sequence is available in GenBank under accession number NC_017586. The structure of ClpP ec in complex with clipibicyclene is available in PDB with accession number 7MK5. Source data are provided with this paper.