Discovery of cahuitamycins as biofilm inhibitors derived from a convergent biosynthetic pathway

Pathogenic microorganisms often have the ability to attach to a surface, building a complex matrix where they colonize to form a biofilm. This cellular superstructure can display increased resistance to antibiotics and cause serious, persistent health problems in humans. Here we describe a high-throughput in vitro screen to identify inhibitors of Acinetobacter baumannii biofilms using a library of natural product extracts derived from marine microbes. Analysis of extracts derived from Streptomyces gandocaensis results in the discovery of three peptidic metabolites (cahuitamycins A–C), with cahuitamycin C being the most effective inhibitor (IC50=14.5 μM). Biosynthesis of cahuitamycin C proceeds via a convergent biosynthetic pathway, with one of the steps apparently being catalysed by an unlinked gene encoding a 6-methylsalicylate synthase. Efforts to assess starter unit diversification through selective mutasynthesis lead to production of unnatural analogues cahuitamycins D and E of increased potency (IC50=8.4 and 10.5 μM).

W idespread antibiotic resistance is currently posing a grave health burden through a multitude of serious infections 1 . The rise in bacterial adaptation can be directly correlated to the paucity of novel classes of antimicrobial agents 2 . In the past few decades, synthetic tailoring has been the primary strategy for enhancing established core scaffolds through analogue generation. Although this approach has been fruitful, no major classes of new antibiotics were introduced between 1962 and 2000 (ref. 3). Therefore, to restore robust access to effective therapeutic agents, it is imperative that we engage in aggressive efforts to discover novel chemical entities with unique microbial targets 2,4 .
Acinetobacter baumannii belongs to the ESKAPE group of major nosocomial opportunistic resistance pathogens (Enterococcus faecalis, Staphylococcus aureus, Klebsiella pneumoniae, A. baumannii, Pseudomonas aeruginosa and Enterobacter sp.), which can spread epidemically among patients causing ventilator-associated pneumonia and bacteremia, with mortality rates as high as 60%, representing a paradigm of pathogenesis, transmission and resistance 5 . Numerous reports have also shown startling emergence of multidrug resistant A. baumannii in hospitals, and also identification of pan-drug-resistant strains at some locations 5,6 . A. baumannii strains possess both intrinsic resistance to antibiotics and a facile ability to acquire genes encoding resistance determinants. In addition, antibiotic resistance of this pathogenic microbe appears to be mediated by their propensity to form biofilms with a highly structured extracellular polymeric matrix, and includes the ability to colonize medical devices. When attached, bacterial cells that comprise the biofilm possess 10-1,000-fold lower susceptibility towards antimicrobial agents compared with planktonic forms 7 . Moreover, despite the central role that bacterial biofilms play during infection, there are currently no drugs specifically targeting biofilms in clinical trials to date 8,9 . Therefore, a precision medicine is urgently required for vulnerable patients to avoid potential life-threatening infections. Recently, developing biofilm inhibitors has become a priority compared with biofilm disruption due to the advantage provided by preventing subsequent dispersion of cells that may have acquired drug resistance [10][11][12] . Although biofilm control by drug targeting has become a high priority objective 7,13 , marine microbes as a source of novel chemical entities remain relatively underexplored 14,15 .
In our continuing effort to identify new structural classes of antibiotics 4 , we employed static-and flow-based high-throughput screening (HTS) assays to survey our natural product extract (NPE) library in the search for new inhibitors of biofilm formation 16 . Here we describe the discovery of three novel secondary metabolites, whose stable production and full structural identification required ribosome engineering, and was facilitated by biosynthetic gene cluster characterization. In addition, we show that the cahuitamycins are derived from two independent starter unit pathways, one of which is genetically unlinked to the core cluster. The convergent pathway enabled us to perform directed pathway engineering to generate a more potent molecule, selectively. Furthermore, mutasynthetic efforts on the ribosomally modulated strain generated two additional novel compounds with enhanced activity against biofilm formation.

Results and Discussion
HTS for biofilm inhibitors against A. baumannii. In an effort to discover new antibiotics to combat hospital-associated infections in patients, we proceeded to screen our marine microbial-derived NPE library to identify biofilm inhibitors of A. baumannii. Natural products account for the majority of currently marketed drugs 17 , and the marine microbiome represents a potential new source of unique chemical entities 18 . We adapted a crystal-violetbased high-throughput assay 19 that was queried against a library of 9,831 marine microbial-derived NPEs to identify the extracts inhibiting biofilm formation as the primary screen. The active extracts were further prioritized by setting the inhibition threshold to 50% followed by a dose-response assay, yielding 31 active NPEs ( Supplementary Figs 1-2). A second round of microbial biological activity analysis was conducted on the top nine most potent extracts ( Supplementary Fig. 2). This study revealed the extract from Streptomyces gandocaensis to be of particular interest due to its ability to inhibit biofilm formation, but showing a limited effect on A. baumannii growth ( Supplementary Figs 1-3).
Ribosome engineering of S. gandocaensis. Following initial identification of the active principles, regrowth of the wild-type S. gandocaensis over several months showed complete loss of production of the active biofilm inhibitor molecules. Therefore, we decided to immediately pursue a ribosome engineering approach to restore and improve production of the active metabolites. This approach has been employed for activation of secondary metabolite production in Streptomyces spp. 20 , and can result in significantly enhanced yields by inducing point mutations in ribosomal protein-encoding genes (for example, rpsL). Several rounds of mutagenesis based on a streptomycin resistance phenotype resulted in an improved strain DHS287 of S. gandocaensis (Supplementary Fig. 4; Supplementary  Table 7), a fourth-generation mutant with restored stable production that generates several-fold increased quantities of active molecules compared with initial wild-type levels. Genetic analysis revealed that the streptomycin-induced ribosome engineering introduced a point mutation in the rpsL gene, which encodes the ribosomal protein S12, in the engineered strain (Supplementary Fig. 5; Supplementary Table 3). Previous studies have shown that mutations in the S12 gene render cells potentially more active for polypeptide synthesis under typical starvation conditions during the late growth phase 20 . This effort appears to be the first reported instance where complete loss of an active, but structurally uncharacterized natural product has been recovered using the ribosome engineering approach.
Cahuitamycin C (3) was also isolated as a white amorphous solid from the same C18 fraction containing 1 and 2. The HRESIMS [M þ H] þ ion peak at m/z 650.2804 provided a molecular formula of C 28 H 39 N 7 O 11 with 13 degrees of unsaturation ( Supplementary Fig. 21). Extensive 1D and 2D NMR analysis indicated that 3 shares structural similarity on much of the carbon backbone compared with 2 (Supplementary Table 1). The only observed difference was localized to the phenyl ring system where an HMBC correlation from d Stereochemical studies for cahuitamycins A-C. Due to the polypeptidic nature of the cahuitamycins, we pursued advanced Marfey's analysis 21,22 to ascertain absolute stereochemistry. Initially, only cahuitamycin B (2) was selected for acid hydrolysis followed by 1-fluoro-2,4-dinitrobenzene-5-alanine amide (FDAA) derivatization. This was based on 2 appearing to be the congener leading to production of cahuitamycins A and C (1 and 3) following cyclization or vice versa (see further below).
A study was conducted to compare m/z 357.27, 382.32 and 400.34 channels from liquid chromatography/electrospray ionization/mass spectrometry (LC-ESI-MS) chromatograms between the L-FDAA and D, L-FDAA-derivatized Ser, Pip and N-OH-Orn (a hydrolysed product of N-OH-N-fOrn) products of 1, respectively. Analysis clearly revealed the absolute configuration of the moieties in the hydrolysate of 1 to be L-Ser, L-Ser, D-Pip and D-N-OH-Orn, respectively ( Fig. 1).
Furthermore, analysis of the Ser, Pip and N-OH-Orn portions of 1 and 3 was conducted as described above for 2, revealing the same stereochemistry. The absolute configuration of the oxazoline ring in 1 and 3 was extrapolated to be D based on similar NMR chemical shifts compared with 2, as well as the absence of an epimerization domain in the first module (CahA) of the cahuitamycin biosynthetic gene cluster (see above, Fig. 2).
Affinity of cahuitamycins towards iron. Cahuitamycins were observed to have siderophore-like properties, and we therefore conducted competition titrations with EDTA to identify their relative iron-binding affinities 23 . In this method, the Fe-cahuitamycin complex was incubated with varying concentrations of EDTA and assessed for Fe distribution using ultraviolet-visible spectroscopy (Methods). The pFe III for cahuitamycins were calculated against the known stability constants for EDTA (pFe III ¼ 23.42) 24 . Three individual titrations were conducted for each natural product, and the representative pFe III values for cahuitamycins 1-3 were measured to be 18.34±0.16, 20.42±0.09 and 17.52±0.12, respectively ( Supplementary Fig. 38). Importantly, the calculated pFe III for 1-3 was significantly lower than those previously reported for hydroxamate siderophores under similar physiological condition (B25-37) 25 . This supports our hypothesis that cahuitamycins possess siderophore properties, but their relatively low affinity may indicate it to be a secondary endogenous function. Interestingly, the diminishing activity of 1 over an extended bioassay time course and the observed inverse proportionality between pFe III of 1-3 against A. baumannii biofilm inhibition could be attributed to this iron chelation property (see below).
Dissecting the cahuitamycin biosynthetic gene cluster. To mine a candidate gene cluster for cahuitamycin biosynthesis, analysis of the draft genome sequence of S. gandocaensis was performed using antiSMASH 26 . As a result, a gene cluster was identified (GeneBank accession code KU363800) as potentially responsible for nonribosomal peptide scaffold biosynthesis, transport and regulation of cahuitamycins A and B (1 and 2; Fig. 2a). Cahuitamycin C (3) biosynthesis was initially expected to involve a S-adenosyl-L-methionine-dependent C-methyltransferase through post-assembly tailoring. However, the absence of genes responsible for the formation of a methyl group at C-23 position of 3 ( Fig. 1) near cluster 1 led to the hypothesis that another genetic locus encoding 6-methyl salicylate synthase (6-MSAS) might be involved in its biosynthesis (Fig. 2). The architecture and annotation of the bifurcated cahuitamycin (cah) gene cluster is shown in Fig. 2 and Supplementary Table 4, respectively. The cluster containing nonribosomal peptide synthetase (NRPS)encoding genes cahA-D, together with genes involved in chain initiation (cahIJ), termination (cahG) and a transcriptional regulator (cahR), is located in a region spanning B40 kb of DNA. Each NRPS protein consists of the essential condensation (C), adenylation (A) and thiolation (T) domains (Fig. 2b), and has a non-colinear architecture.
CahA is proposed to catalyse oxazoline ring formation together with a putative salicylate synthase CahI and salicylate-AMP ligase CahJ (Fig. 2b). CahI is homologous to MbtI from Mycobacterium tuberculosis and is predicted to convert chorismate to salicylate ( Fig. 2b) 27 . CahJ bears high similarity to a salicylate-AMP ligase MxcE from Sorangium cellulosum So ce56 (ref. 28), which suggests that it activates salicylate by adenylation (Fig. 2b). The identity of amino acid building blocks selected and activated by the Cah NRPS were analysed by examining the specificityconferring residues in each A domain (Supplementary Table 5) 29 . The A domain in CahA is predicted to recognize L-Cys (Supplementary Table 5), though only L-Ser is loaded according to the structure of cahuitamycins ( Fig. 1), which is also observed in the amychelin biosynthetic system 30 . The heterocyclization between Ser and the carbonyl group to form an oxazoline ring scaffold can be attributed to the Cy domain in CahA, and the resulting peptide product is then transferred to CahB for synthesis of 1 and 3. Similar to formation of gobichelins, where the oxazoline ring of gobichelin A is hydrolysed into a linear form under non-acidic condition, 1 may convert to 2 with a free Ser through hydrolysis of the oxazoline ring 31 . The in silico analysis of A domains in CahB predicts activation of L-Ser and L-N-OH-N-fOrn by CahB-A 1 and CahB-A 2 , respectively (Supplementary Table 5). Subsequent to the incorporation of L-Ser by CahB-A 1 , a L-N-OH-N-fOrn moiety, which is synthesized from L-Orn by a putative Orn hydroxylase (CahMO) and a putative N-formyltransferase (CahFT) 32 , is appended to the growing chain (Supplementary Table 4). The L-N-OH-N-fOrn moiety linked to the T 2 domain of CahB undergoes epimerization by the epimerase (E) domain 33 , which is consistent with the stereochemistry of cahuitamycins ( Fig. 1; Supplementary Figs 8 -37). Biosynthesis of cahuitamycins likely involves formation and loading of the piperazic acid (Pip) building block into the growing peptide chain. Although the relatively rare Pip moiety is found in several biologically active secondary metabolites [34][35][36][37][38][39] , the precise biosynthetic events for Pip elaboration have not been elucidated. Recent precursor incorporation experiments have suggested that generation of the Pip moiety in kutzneride occurs before incorporation into the respective peptide chain 35 , similar to the formation of a hydrazo linkage in valanimycin 34 . It is likely that biosynthesis of the Pip moiety in cahuitamycin is initiated by Nhydroxylation of the amino group of Orn by a putative CahMO. The resulting L-N-OH-Orn residue attached to the T domain of CahC would be converted to its corresponding D-stereoisomer (D-N-OH-Orn) by the E domain within the same module followed by nucleophilic attack of the amino group to close the ring. Alternatively, since the A domain in the CahC is homologous to L-glutamine (Gln)-activating A domains (Supplementary Table 5), it is possible that Gln may be incorporated as a branch point from primary metabolism into the piperazate biosynthetic pathway (Fig. 2) 40,41 .
Further in silico analysis of the A domain in CahD (Supplementary Table 5) predicted loading of b-Ala onto the T domain consistent with the structure of cahuitamycins. Moreover, it further affirms that the presence of b-Ala residue is likely formed from L-aspartate (L-Asp) by the cahF gene product 42 . The CahD polypeptide contains neither C-terminal thioesterase nor a reductase domain usually required for chain release in nonribosomal peptide biosynthesis, and likely involves a specific hydrolase for release of the fully assembled peptide chain from the T domain of the last module CahD. We propose that following final extension by b-Ala, CahG (a/b hydrolase superfamily) catalyses in trans release of the peptide chain through hydrolysis of the T domain-bound thioester as described previously in coelichelin biosynthesis 43 .
Moreover, CahE, an MbtH-like protein homologue, likely contributes to the stimulation of A domains in the Cah NRPS 44 . CahH, which belongs to the major facilitator superfamily, is similar to EntS, an enterobactin efflux exporter of Escherichia coli. Proteins encoded by CahT1 through CahT8 are putatively required for iron uptake and translocation of cahuitamycin across the cell membrane (Supplementary Table 4). Therefore, these proteins might be involved in the export of cahuitamycins from the cytoplasm into the medium. Open reading frame 1 (ORF1), ORF2 and ORF3 show sequence similarity to oxidoreductase, acyltransferase and 6-aminohexanoate-oligomer hydrolase, respectively, and their relationship to cahuitamycin biosynthesis remains unclear (Supplementary Table 4).
Biosynthesis of cahuitamycin C. We next sought to confirm the function of the assigned biosynthetic cluster by deletion of cahI in S. gandocaensis through insertion of the kanamycin resistance gene (aph). Surprisingly, the DcahI S. gandocaensis strain DHS334 (Methods; Supplementary Fig. 7; Supplementary  Table 7) gave exclusive production of 3, while the production of 1 and 2 were completely abolished (Fig. 3d). The result indicated a possible unlinked source involved in providing 6methylsalicylate starter unit. A BLAST analysis using bacterial 6-MSASs including various putative candidates from other bacterial secondary metabolite gene clusters [45][46][47][48][49] , led to the identification of a single iterative type I PKS gene encoding a 6-MSAS (genetic locus 2; GeneBank accession code KU363801) located at least B50 kb away from cluster 1 based on our S. gandocaensis draft genome-sequencing data (Fig. 2a). This finding led us to propose that the 6-methylsalicylate starter unit is likely derived from the 6-MSAS identified from bioinformatics mining, which is subsequently adenylated by CahJ and loaded onto the T 1 domain of CahA for production of 3 (Fig. 2). As in other bacterial PKSs in its class, the putative cahuitamycin 6-MSAS (genetic locus 2; cah MSAS) possesses keto synthase, acyltransferase, dehydratase, ketoreductase and acyl carrier protein domains on a single protein (Supplementary Table 4). The hypothesis was further substantiated by the dosing effect observed by introducing exogenous 6-methylsalicylic acid into cultures of the DcahI S. gandocaensis strain. The feeding study provided increased production of 3, while no changes were observed in production of 1 and 2 ( Supplementary Fig. 39). This result, together with the lack of other potential sources of 6methylsalicyclate, or a candidate methyltransferase encoded within the S. gandocaensis genome appears to rule out the possibility of any posttranslational modification of 1 to produce 3. Furthermore, production of 1 and 2 was restored by external supplementation of salicylic acid to the DcahI growth medium (Fig. 3e), representing chemical complementation and confirmation of its role in the production of cahuitamycins. These studies substantiated that biosynthesis of 3 is independent of cahI-mediated salicylic acid biosynthesis, and more importantly cahuitamycin C (3) appears to be derived from a convergent biosynthetic pathway (Fig. 2). Secondary metabolites assembled from unlinked gene clusters have been reported in previous studies 32,50,51 , but to the best of our knowledge this is the first example of a starter unit pathway that is genetically unlinked to a corresponding bacterial NRPS. The observation provides an intriguing example where S. gandocaensis may be diversifying its biosynthetic machinery to produce a more potent/ toxic metabolite for enhanced competitiveness in its natural habitat.
Mutasynthetic generation of cahuitamycin analogues. The biosynthetic process for generating 3 revealed the possible tolerance of CahJ towards structurally related salicylic acid substrates, and a promising avenue for producing new cahuitamycin analogues by mutasynthesis 52 . In this approach, we provided the DHS334 DcahI strain with a series of substituted benzoic acid substrates ( Supplementary Fig. 40), and sequentially assessed their incorporation to make new analogues (Fig. 4).
Of the series of unnatural starter units tested, feeding of methyl-hydroxybenzoic acid substrates showed some level of incorporation. However, only assimilation of 5-methylsalicylic acid into the DcahI pathway provided isolable quantities of the new analogue, cahuitamycin D (4; Fig. 4a; Supplementary  Fig. 40). Cahuitamycin D (4) was isolated by reverse-phase HPLC (RP-HPLC) from crude extract of DcahI mutant. The HRESIMS [M þ H] þ ion peak at m/z 650.2706 provided similar molecular formula as of 3, C 28 H 39 N 7 O 11 . Extensive 1D and 2D NMR data were acquired for 4, which indicated the expected structural similarity with 3, as per the mutasynthesis hypothesis ( Fig. 4a; Supplementary Table 2). The structure showed a similar carbon backbone with a clear difference at the phenyl ring system compared with 3. Furthermore, during isolation of 4, we observed another HPLC peak eluting with this natural product (Fig. 4b). Interestingly, the HRESIMS [M þ H] þ ion peak at m/z 668.2830 suggested the same backbone with likely additional hydration. The new molecule cahuitamycin E (5) was shown by 1D and 2D NMR analysis to share structural similarity on much of the carbon backbone compared with 4 (Supplementary Table 2). The only plausible difference could be traced to the C-5 position where Pip moiety in 4 was substituted with L-N-OH-Orn in 5, suggested based on the COSY relay observed from d H 4.33 (H-5) to d H 3.45 (H-8) along with absence of any significant HMBC correlation from H-8 to C-9 (d C 172.2) (Fig. 4b). The observation was compelling as it indicates that either the cyclization to form Pip moiety is occurring after the insertion of L-N-OH-Orn in 5 or, alternatively, the domain is capable of accepting both moieties separately. In addition, the study also indicated that L-glutamine might not be a substrate for the CahC A domain during biosynthesis of cahuitamycins.
The absolute stereochemical configuration of the cahuitamycins D and E (4 and 5; Fig. 4) was extrapolated to be the same as cahuitamycins A-C (1-3) based on very similar chemical shifts observed for all nuclei obtained through detailed 1D and 2D NMR analysis (Supplementary Table 2).
Biological activity associated with cahuitamycins A-E. Cahuitamycins A-C (1-3) were next tested for their ability to inhibit biofilm formation of A. baumannii using a high-resolution optical secondary assay system. The primary high-throughput static biofilm assay leading to identification of 1-3 was conducted using crystal violet staining followed by optical density measurements (Fig. 5a,b). The result from this assay showed that 1 was able to inhibit, while 2 had no effects on bacterial growth or biofilm formation (Fig. 5a,b). Interestingly, 3 possessed the highest potency with half-maximal inhibitory concentration (IC 50 ) values of 14.5 mM while having a limited effect on growth of A. baumannii. In addition, a confocal laser scanning microscope imaging analysis of biofilm formation and biomass was performed (Methods). In this assay, the biofilm was developed on a 12-well glass plate with solvent (no compound) as a negative control and only 240 nM of 1 and 3 as treated. After staining with fluorescent dye SYTO9 and propidium iodide, the biofilm structure was observed (Fig. 5c). The reconstructed threedimensional (3D) biofilm images obtained showed a much thinner biofilm and also significantly less total biomass for the glass plates treated with 1 and 3 compared with control (Fig. 5c). Furthermore, the glass plate treated with 3 showed a much lower level of total biomass compared with 1 at the same concentration (0.10 mm 3 versus 0.18 mm 3 ), which confirmed 3 as the more potent biofilm inhibitor (Fig. 5c).This finding indicates that cahuitamycins prevent biofilm formation through a specific, yet currently undefined biochemical target with negligible impact on cell growth (Fig. 5a,b). Furthermore, cahuitamycin C (3) was tested against the growth of A. baumannii separately ( Supplementary Fig. 41), and revealed an IC 50 value of Z1.0 mM ( Supplementary Fig. 42). In addition, when 1 (the most abundant molecule) was tested in iron-complexed form (owing to its moderate iron chelation affinity), it showed minimal impact on both growth and biofilm formation of the pathogen (Fig. 5a,b). To probe iron chelation as a possible mode of action for the cahuitamycins, a known siderophore desferrioxamine was used as control, which showed no activity, suggesting that iron chelation (or depletion) may not be the basis for cahuitamycin activity ( Supplementary Figs 38 and 44). However, loss of inhibition of biofilm formation over time could be attributed to the metal-complexed cahuitamycins ( Supplementary Figs 38 and 44), as the test media is replete with iron and 1 in complex with Fe III has not displayed comparable inhibitory activity (Fig. 5a). This observation is consistent with previous studies, which have reported that clinical isolates of A. baumannii are able to grow in the presence of high concentrations of different iron chelators due to its ability to express a high-affinity extracellular siderophore, acinetobactin 53,54 . In addition, the structural differences between cahuitamycins with previously reported amychelin 30 , results in a significant change in iron affinities, which can be further explored for biofilm inhibitory properties in future studies. Next, we sought to probe the specific stage and extent of inhibition during biofilm formation employing a secondary flow cell assay (Supplementary Fig. 44; Methods). This analysis revealed almost the same number of cells initially attached to the surface before starting treatment. After 6 h of inhibitor supplementation, the flow cells treated with 1 showed only a few microcolonies of the pathogen, while the control (no treatment) and desferrioxamine showed formation of a robust biofilm. After 24 h of supplementation, the control and desferrioxamine biofilm became thicker with the flow cells exposed to 1 showing larger microcolonies and initiation of biofilm formation by A. baumannii.
As described above, the mutasynthetic study on S. gandocaensis strain DHS334 led to the isolation of two additional analogues cahuitamycins D-E (4-5). The new molecules were also subjected to static biofilm assays and the result demonstrated that 4 displayed almost twice the potency (IC 50 ¼ 8.4 mM) compared with 3, which we earlier considered to be the most active molecule ( Fig. 6; Supplementary Figs 41-43). Furthermore, cahuitamycin E (5) also displayed significant activity with an IC 50 of 10.5 mM (Fig. 6; Supplementary Figs 41-43). In addition, compounds 3 and 4 were also tested for disruption of established A. baumannii biofilm revealing low activity with an IC 50 of 692 and 535 mM, respectively (Supplementary Fig. 45; Methods), establishing their better efficacy as biofilm formation inhibitors. Although prophylactic antibiotic administration preceding surgery is routinely successful in reducing infection rates, it has little or no protective effects in surgical procedures involving implants or prostheses 55 . Therefore, catheters and implants coated with biofilm-inhibiting antimicrobial agents are urgently needed as an effective, precision therapeutic option. Our current data indicate that a key pharmacophore of the cahuitamycins is likely the 2-hydroxybenzoyl-oxazoline group where relatively minor modifications can result in an increase (as in case of 3-5) or decrease (as for 2) of anti-biofilm activity, with the potential for further structure activity relationship studies using synthetic chemistry or Cah pathway engineering. The application of the streptomycin resistance-mediated screening of the Streptomyces strain identified from HTS against A. baumannii has yielded a novel structural class of biofilm inhibitors derived from a marine microbial NPE library. Cahuitamycins A-E (1-5) inhibit the ability of A. baumannii to generate a biofilm, which is involved in the emergence of a widespread antibiotic resistance phenotype 6,56 . The cahuitamycins are most active in apo form, with chelation of Fe III over time resulting in attenuated biological activity. The cahuitamycins along with allied chemical entities with distinct biological activity provides a primary foundation for future medicinal chemistry and synthetic exploration towards the development of an efficacious drug to prevent or limit biofilm formation.
We have also demonstrated that production of cahuitamycins A-C is specified by a convergent pathway specified by the cah biosynthetic gene cluster comprised of two unlinked genetic loci. The inherently flexible starter unit adenylating enzyme CahJ was exploited to produce two additional analogues in the engineered DcahI strain DHS334. The new cahuitamycin D analogue displays twofold-enhanced biofilm inhibitory activity compared with the cahuitamycin C natural product. These results establish a unique opportunity for developing and discovering new antibiotics from genetically engineered strains bearing inherent flexibility in pathway initiation processes. More importantly, given the simultaneous decline in antibiotic drug discovery and increased incidence of multidrug resistant bacteria, the cahuitamycins may represent a propitious starting point for discovery and development of new therapeutics against dangerous human pathogens involving biofilm formation. The procedure for the isolation of actinomycetes from these samples was previously described by Magarvey et al. 16 Maintenance and propagation of cultures were performed using standard media and protocols where 500 mg of wet sediment was diluted in 10 ml of sterile water and vortexed for 10 min. Then, 1 ml of this suspension was applied directly to the top of the discontinuous sucrose gradient and centrifuged for 30 min at 300g. A volume of 500 ml of the 20, 30 and 40% layers was then plated to HVA agar supplemented with 10 mg ml À 1 chlortetracycline, 25 mg ml À 1 cyclohexamide and 25 mg ml À 1 of nalidixic acid. The plates were then incubated at 28°C for 1 month. The colony was picked off the plate and streaked onto ISP2 agar until pure. Seed cultures were grown in 17-ml dual-position cap tubes containing 2 ml of ISP2 and grown for 4 days on a rotary shaker at 200 r.p.m. The seed culture was then poured into a 250-ml baffled flask containing 100 ml of ISP2 and grown for 18 days on a rotary shaker at 200 r.p.m. The culture was centrifuged at 4,000 r.p.m. for 10 min to remove the cells, and 2 g of XAD16 resin (Sigma-Aldrich, St Louis, Mo) contained within a polypropylene mesh bag was added to the broth and incubated overnight on the rotary shaker. The resin bag was removed and placed into 10 ml of MeOH, followed by 10 ml of acetone and 10 ml of ethyl acetate (EtOAc). Each of the three fractions was dried in vacuo and reconstituted to a final concentration of 15 mg ml À 1 in DMSO.
Culture maintenance and fermentation. Seed cultures of 100 ml (5 Â ) of ISP2 media (1% malt extract, 0.4% yeast extract, 0.4% dextrose and 3% NaCl) were inoculated with a loop-full of vegetative cells from an oatmeal plate (6% oat meal, 1.25% agar and 3% NaCl) culture of S. gandacoensis and incubated with shaking (200 r.p.m.) at 28°C for 5 days. A 25 ml portion of the seed cultures were transferred to a 2.8-l Fernbach flask containing 1.5 l of the ISP2 medium, and the 39 l fermentation was carried out on a rotary shaker (200 r.p.m.) at 28°C for 18 days. After 14-18 days of growth, the cultures were collected by centrifugation. The resulting cell-free broth was subjected to solid phase extraction using 15 g of Amberlite XAD-16. The resin was then separated by filtration and subjected to organic extraction using MeOH:EtOAc (1:1).
Natural product extract library. The NPE library at the University of Michigan Center for Chemical Genomics contained B20,000 extracts. Each extract in the library is derived from marine samples collected from all over the world, including Costa Rica, Panama and Papua New Guinea. Some of these samples are from isolated microbes (n ¼ 19,055), while others were derived from field-collected biomass samples ('macrosamples,' n ¼ 800). Previous work describes in detail how these extracts are prepared for the library 16 .
PCR amplification, cloning and sequencing of 16S rDNA. Genomic DNA of S. gandocaensis was isolated using the Wizard Genomic DNA Purification Kit according to the manufacturer's instruction. The 16S ribosomal DNA (16S rDNA) gene was amplified by PCR with the universal primers FC27 and RC1492 (Supplementary Table 6) 57 using the genomic DNA as a template. The PCR fragments were cloned into pGEM-T Easy vector (Promega) and the resulting plasmid containing 16S rDNA of S. gandocaensis was sequenced using T7 and SP6 primers. Phylogenetic analysis of 16S rDNA of S. gandocaensis. Phylogenetic analyses were conducted using GENEIOUS R6 that is available from http://www. geneious.com/ (ref. 58). The evolutionary history was inferred using the Neighbour-Joining method. The bootstrap consensus tree inferred from 500 replicates was taken to represent the evolutionary history of the taxa analysed ( Supplementary Fig. 6).
Generation of ribosome-engineered mutants of S. gandocaensis. Spores (10 8 -10 9 ) of wild-type S. gandocaensis were spread on R2YE agar containing various concentrations of streptomycin, followed by 15 days incubation at 28°C to allow the development of streptomycin-resistant colonies. Eighty resistant colonies ( Supplementary Fig. 4) were obtained from the plate containing a high level (10, 50 and 100 mg ml À 1 ) of streptomycin and grew in ISP2 media with streptomycin for 14 days at 28°C. This step was repeated four times to obtain spontaneous streptomycin-resistant mutants. The ability of selected mutants to produce cahuitamycin was tested by LC-MS using Agilent extend C18 5 mm 2.1 Â 150-mm column on gradient condition 30-85% acetonitrile (w/0.1% formic acid) in H 2 O (w/0.1% formic acid) over 15 min with additional 5 min of equilibration time.
Amplification and sequencing of rpsL. Nine strains, including the parent strain and eight high-yielding recombinants, were selected to determine mutations in the rpsL gene in this study. The rpsL gene was amplified by PCR with primers SR185 and SR186 (Supplementary Table 6), using genomic DNA as a template and cloned into pGEM-T easy vector (Promega) and sequenced. Mutations in the wild-type rpsL gene were determined by DNA sequencing 59 (Supplementary Fig. 5; Supplementary Table 3).
Isolation and purification of cahuitamycins. The organic extracts obtained from the engineered strains were concentrated under vacuum to afford the crude extracts (B350 mg) obtained from 100-ml culture. The crude extract was assayed in the developed in vitro crystal violet at 10 and 1.0 p.p.m. The bioactive extract was then further purified by RP-HPLC on a gradient of 10-75% ACN and was followed by ultraviolet-visible photodiode array detection at 215 nm to yield semi-pure compounds 1 (6.7 mg), 2 (2.4 mg) and 3 (2.9 mg). Compounds were again subjected to repurification over RP-HPLC on isocratic condition of 35% MeOH (0.1% FA) using C-8 column to get compounds 1 (5.1 mg), 2 (1.1 mg) and 3 (1.6 mg). Generation of cahI deletion mutant. The cahI gene was inactivated by an insertional inactivation via double-crossover homologous recombination ( Supplementary Fig. 7A). A knock-out plasmid pSRP50 based on pKC1139 (ref. 60) was constructed by amplifying kanamycin resistance gene as a selection marker from plasmid pYJ276 (ref. 61), and left-and-right-flanking regions of the cahI gene using the genomic DNA of S. gandocaensis as a template. The primer pairs SR144-SR145, SR146-SR147 and SR148-SR149 (Supplementary Table 6) were designed for the amplification of left-and-right-flaking fragments of cahI gene, and selection marker, respectively. DNA assembly was performed using Gibson assembly 62 master mix (New England Biolabs) according to the manufacturer's instructions. The plasmid pSRP50 was passaged through methylation-deficient E. coli ET12567 and then introduced into the S. gandocaensis by protoplasts-based transformation 63 . The target region of cahI gene was then disrupted by an insertional inactivation via double-crossover homologous recombination. The desired mutant DcahI was selected by its kanamycin-resistant and apramycinsensitive phenotype (Supplementary Fig. 7b) and verified by PCR ( Supplementary  Fig. 7c) using the primer pair SR233-SR234 (Supplementary Table 6). The resulting cahI deletion mutant of S. gandocaensis was designated as DHS334.
Chemical complementation of mutant DHS334 and mutasynthesis. Mutant strain of DHS334 was first precultured in 3 ml R2YE liquid medium for 15 days at 28°C and then 3 ml of the seed culture was used to inoculate 100 ml of the same medium, followed by cultivation for 15 days at 28°C. Salicylic acid and a series of substituted benzoic acid substrates, 2-hydroxybenzoic acid, 2,3-dihydroxybenzoic acid, 2,4-dihydroxybenzoic acid, 2-fluorobenzoic acid, 2-hydoxy-5-methylbenzoic acid (5-methylsalicylic acid) and 2-hydoxy-6-methylbenzoic acid (6-methylsalicylic acid) were added every alternate day to separate 100-ml cultures of DHS334 at a final concentration of 500 mM for 15 days. The products were first extracted using Amberlite XAD-16. The resin was separated and subjected to organic extraction using MeOH:EtOAc (1:1) for LC-MS analysis as described above.
The substrate-fed crude extract was then further purified by RP-HPLC on a gradient of 10-75% ACN and was followed by ultraviolet-visible photodiode array detection at 215 nm to yield semi-pure compounds 4 (2.4 mg) and 5 (2.6 mg). Compounds were again subjected to repurification over RP-HPLC on isocratic condition of 35% MeOH (0.1% FA) using C-8 column to get compounds 4 (1.1 mg) and 5 (1.3 mg).
Cahuitamycin D (4 64 . Briefly, purified Fe-cahuitamycins, prepared as described above, was dissolved in HEPES buffer (10 mM HEPES and 0.1 M KCl, pH 7.4) and five different concentration ranges of ETDA were added from EDTA stock solutions also prepared in HEPES buffer. Each reaction consisted of a total volume of 0.25 ml, a final concentration of 0.1 mM Fe-cahuitamycins and a range of 5-8,000-fold EDTA (relative to Fe-cahuitamycins) in HEPES buffer. The reaction was allowed to equilibrate at room temperature for at least 24 h, and ultraviolet-visible spectra were subsequently recorded. The composite spectra contain contributions from both Fe-EDTA and Fe-cahuitamycins. The e of both species as a function of l were determined in HEPES buffer (see Supplementary Fig. 38 for Fe-cahuitamycins) and used to deconvolute the spectra. The contribution of Fe-cahuitamycins was subtracted from the composite spectra using the 435 nm absorption band, and the e of both  (1), which has been derived by Abergel et al. 64 . All the data sets are shown in Supplementary Fig. 38, yielding a pFe III as reported in original article.
Assay development and high-throughput screening assay. A static biofilm assay was developed from a previously reported method 19 . The biofilm assay measures the adhesion of bacteria to the surface of polystyrene 384-well microtiter plates. Adherent cells are detected by staining with crystal violet and subsequent washing to remove nonadherent planktonic cells. Inhibitors of biofilm formation prevent the bacteria from adhering to the surface of the microtiter plate and reduce the amount of crystal violet retained after washing. Assay quantification was performed by OD 600 absorbance measurement of the crystal-violet-stained biofilm. The assay was optimized for plate surface treatment, time, temperature, media, media concentration, crystal violet concentration, wash method, inhibitor sensitivity and culture inoculum preparation. The A. baumannii test strain (ATCC 17978) was maintained as a frozen stock at À 80°C in 20% glycerol. A volume of 5 ml of Mueller Hinton II cation adjusted media (Becton Dickinson, catalogue no. 212322) was inoculated and grown at 37°C, 180 r.p.m. shaking for 4-6 h. The culture was then diluted 1:50 and incubated an additional 2 h. The resulting cells were washed four times by centrifugation and resuspension of the pellet in media was performed to remove any metabolites or cell signalling factors. The assay inoculum was prepared by diluting the cells to 0.008 OD 600 in 10% Mueller Hinton II cation adjusted media (diluted in 18 Ohm deionized water). The assay plates (Corning, catalogue no. 3,680; 384-well non-treated polystyrene) were prepared by dispensing 20 ml of 10% Mueller Hinton II cation adjusted media into columns 1-22 using a Multidrop Combi dispenser (Thermo Fisher). The NPE samples were added as 0.2 ml of 15 mg ml À 1 stocks in DMSO using a Biomek FX high-density pintool (Beckman Coulter). Column 1 and 2 contained DMSO without NPEs and served as the negative control. The positive control was added to column 23 and 24 as 20 ml of 40 mM Baicalein (Sigma-Aldrich, catalogue no. 11712). The bacterial cells were added to the entire plate in 20 ml of the prepared inoculum using a Multidrop Combi dispenser (Thermo Fisher). The resulting assay contained 40 ml of 10% Mueller Hinton II cation adjusted media with 0.004 OD 600 bacterial cells, 0.5% DMSO and 75 mg ml À 1 NPE. The assay plates were incubated at 30°C for 20 h, stationary, in a humidified incubator. The biofilm was stained by adding 10 ml of filtered 1% crystal violet and incubating for 30 min at room temperature. Excess crystal violet and nonadherent planktonic cells were removed by washing three times with 150 ml of PBS using an ELX405 plate washer (Bio-tek). The wash program used a low velocity dispense rate and an aspiration height of 3.8 mm; B10 ml of PBS remained in the well after washing. The Crystal-Violet-stained biofilm was solubilized by the addition of 50 ml of 100% ethanol and allowed to develop overnight. Quantification was performed by measuring the OD 600 absorbance using a Pherastar plate reader (BMG).
3D imaging of biofilm development. For visualizing bacterial biofilm formation, the same method as described previously was used for biofilm formation but in  65 . After 20 h of incubation at 30°C, the supernatant was removed, and the well surface was rinsed briefly with 1 Â PBS buffer three times. PBS buffer (100 ml) containing 1 mM of SYTO 9 and 10 mM of propidium iodide was added into the well. The plate was incubated for 15 min in dark. Fluorescent images were acquired with an Olympus Fluoview FV1000 confocal microscope (Olympus, Markham, Ontario) with Melles Griot Laser supply and detectors and filter sets for monitoring SYTO 9 and propidium iodide fluorescence. Images were obtained using an oil immersion Â 60 objective lens. 3D images were reconstructed using the Amira software package (Amira, San Diego, CA) from a stack of sectional images of biofilm samples. Biofilm biomass and per cent coverage were calculated based on those images using software COMSTAT 65 .
Micrograph of biofilm in flow cells. Biofilms were formed by A. baumannii ATCC 17978 in flow chambers at room temperature. Images were taken at 1, 6 and 24 h (using a Â 60 lens) after inoculation. A flow cell biofilm was achieved by using a flow cell chamber (ACCFL0001, Life Science Incorporated, Greensboro, NC). Briefly, overnight cultures of A. baumannii ATCC 17978 grown in MHII broth at 37°C were diluted 100-fold with fresh MHII broth. A volume of 1 ml of dilution was injected into a flow cell chamber and allowed to settle 1 h for attachment of bacteria, followed by initiation media flow at a flow rate of 4 ml À 1 h À 1 . Flow media was 10% MHII broth (control), 10% MHII broth with 7.5 mg ml À 1 of desferoxamine or 10% MHII broth with 7.5 mg À 1 ml of cahuitamycin A. Micrographs of the biofilm were acquired at 1, 6 and 24 h after the start of flow media using an Olympus IX70 microscope with a Â 60 lens. Flow was still running while acquiring the micrograph.
Minimum inhibitory concentration test. Cultures of A. baumannii ATCC 17978 grown overnight were diluted into fresh 10% Mueller Hinton II broth to get 10 5 colony-forming unit per ml by determining optical density at OD 600 . A volume of 100 ml of diluted cultures were incubated in individual wells of 96-well plate containing 1% of DMSO and various concentrations of cahuitamycins C (3) and D (4). The plate was incubated at 30°C with shaking at 150 r.p.m. The optical density of each well was monitored every 1 up to 24 h using a microtiter plate reader (Synergy HT, Bio-Tek, Winooski, VT). The number of viable bacterial cells was determined by counting colony-forming unit.
Biofilm dispersal assay. Biofilm dispersal assay was performed as previously reported 66 . Briefly, A. baumannii ATCC 17978 biofilm was developed as described in minimum inhibitory concentration test. After 16 h incubation at 30°C, the suspension in each well was removed by pipetting gently. A volume of 100 ml of 1 Â PBS buffer containing cahuitamycins C (3) and D (4) at various concentrations was placed into each well. The plate was incubated at 30°C for 12 h to allow cahuitamycins to disassemble pre-formed biofilms. Then, the suspension was removed by pipetting. The remaining attached biofilm was quantified by crystal violet assay.