Genomics‐Driven Discovery of a Novel Glutarimide Antibiotic from Burkholderia gladioli Reveals an Unusual Polyketide Synthase Chain Release Mechanism

Abstract A gene cluster encoding a cryptic trans‐acyl transferase polyketide synthase (PKS) was identified in the genomes of Burkholderia gladioli BCC0238 and BCC1622, both isolated from the lungs of cystic fibrosis patients. Bioinfomatics analyses indicated the PKS assembles a novel member of the glutarimide class of antibiotics, hitherto only isolated from Streptomyces species. Screening of a range of growth parameters led to the identification of gladiostatin, the metabolic product of the PKS. NMR spectroscopic analysis revealed that gladiostatin, which has promising activity against several human cancer cell lines and inhibits tumor cell migration, contains an unusual 2‐acyl‐4‐hydroxy‐3‐methylbutenolide in addition to the glutarimide pharmacophore. An AfsA‐like domain at the C‐terminus of the PKS was shown to catalyze condensation of 3‐ketothioesters with dihydroxyacetone phosphate, thus indicating it plays a key role in polyketide chain release and butenolide formation.


Introduction
Thec onstant competition between microbes and their environment has driven the evolution of specialised metabolite production in bacteria, enabling rapid ecological adaptation. [1] Such metabolites frequently find important applications in medicine,a sa ntibiotics,a nticancer agents and immune modulators,a nd agriculture,a si nsecticides,h erbicides and fungicides.Gram-negative bacteria belonging to the Burkholderia genus produce aw ide array of bioactive specialised metabolites,i ncluding the respiratory toxin bongkrekic acid, anti-proliferative agents such as thailanstatin and spliceostatin, and the antibiotics enacyloxin IIa and gladiolin. [2][3][4][5][6] Despite their structural complexity and diversity, these molecules are biosynthesised from simple building blocks by modular polyketide synthase (PKS) and nonribosomal peptide synthetase (NRPS) assembly lines,o ften harbouring non-canonical characteristics. [7,8] Recently,w e have shown that the opportunistic pathogen Burkholderia gladioli BCC0238, isolated from the lung of ac ystic fibrosis (CF) patient, produces ar ange of specialised metabolites, dependent on the carbon source.When glycerol is used as the carbon source,t his strain produces gladiolin, an ovel macrolide with promising activity against Mycobacterium tuberculosis,a nd the swarming inhibitor icosalide A1, which was originally isolated from af ilamentous fungus,b ut has subsequently been shown to originate from a Burkholderia symbiont. [6,9,10] Switching to am ixture of glycerol and ribose as carbon sources,i nduces the production of bolagladins A and B, novel lipodepsipeptides containing au nique citrateprimed fatty acid and an unusual dehydro-b-alanine residue. [11] Although several specialised metabolites and their associated biosynthetic gene clusters (BGCs) have already been identified in B. gladioli BCC0238, analysis of the complete genome sequence of this strain indicated that it contains numerous cryptic BGCs,the metabolic products of which are currently unknown. [6] Here we report the discovery of gladiostatin, an ovel member of the glutarimide class of polyketide antibiotics,a st he metabolic product of ac ryptic trans-acylt ransferase (trans-AT) PKS-encoding BGC in B. gladioli BCC0238. This unusual metabolite,w hich contains arare 2-acyl-4-hydroxy-3-methylbutenolide in addition to the 2, 6-piperidinedione common to all glutarimides,i sa ctive against yeast and has promising activity against several human cancer cell lines.G ene disruption experiments confirmed that the BGC directs the biosynthesis of gladiostatin, and chain release from the PKS was reconstituted in vitro, providing insights into the mechanism for formation of the unusual butenolide moiety.T hese experiments enabled us to propose ab iosynthetic pathway for gladiostatin, the first glutarimide to be isolated from Gram-negative bacteria, illuminating the role played by horizontal gene transfer in trans-ATP KS evolution.

Results and Discussion
Genome Mining Identifies aC ryptic trans-ATP KS Predicted to Assemble aNovel Glutarimide Building on our previous work in B. gladioli BCC0238, in silico analysis of the genome sequence revealed a % 50 kb cryptic BGC encoding at rans-ATP KS on the second chromosomal replicon ( Figure S1). Bioinformatics analyses indicated that several of the proteins encoded by this BGC are very similar to enzymes known to be involved in the biosynthesis of the glutarimide class of polyketide antibiotics in Streptomyces species (Figure 1a and b), including cycloheximide (1), [12,13] 9-methylstreptimidone (2), [14,15] migrastatin/iso-migrastatin (3) [16][17][18] and lactimidomycin (4;F igure 1c, Figure S2). [19,20] Glutarimides have potent antifungal activity and inhibit eukaryotic translation by blocking the binding of tRNAt ot he E-site of the 60S ribosomal subunit. [21,22] They also possess promising anticancer activity and several members of the family have been reported to inhibit tumour cellmigration. [23,24] Three conserved genes (smdFGH, chxBCD, ltmHCD and mgsHCD in the 9-methylstreptimidone,c ycloheximide,lactimidomycin and migrastatin BGCs,respectively) are proposed to be involved in the creation of ac ommon malonamyl thioester starter unit for the PKSs that assembles these metabolites ( Figure 1a). [12,[15][16][17][18][19][20] These genes encode:a n acyltransferase (AT; SmdF/ChxB/LtmH/MgsH) that is proposed to malonylate an acyl carrier protein (ACP;S mdG/ ChxC/LtmC/MgsC);a nd an asparagine synthetase homologue (SmdH/ChxD/LtmD/MgsD), which is hypothesised to convert the resulting malonyl thioester to am alonamyl thioester. Thefirst module in each PKS (located in the SmdI, ChxE, LtmE and MgsE subunits) contains seven conserved domains that are proposed to elaborate the malonamyl starter unit into acommon 2-(2,6-dioxopiperidin-4-yl)acetyl thioester intermediate (Figure 1b). [25] The gdsB, gdsC and gdsD genes in the cryptic B. gladioli BGC encode homologues of the three proteins hypothesised to create the malonamyl thioester starter unit ( Figure 1b). Moreover,the first module of the PKS encoded by this cluster (situated in the GdsE subunit) has an identical seven-domain architecture to the corresponding modules of the PKSs that assemble the Streptomyces glutarimides.H owever,t he domain architecture of subsequent modules in the B. gladioli PKS differs significantly from the Streptomyces glutarimide assembly lines and aunique AfsA-like domain (pfam03756identified by ac onserved domain search) is appended to the C-terminus of the final PKS module (Figure 1b). AfsA catalyses the condensation of dihydroxyacetone phosphate (DHAP) with a b-keto thioester to form ap hosphorylated butenolide intermediate in the biosynthesis of A-factor, as ignalling molecule that controls morphological differentiation and antibiotic production in Streptomyces griseus (Figure 1d). [26] AfsA homologues are proposed to catalyse analogous reactions in the biosynthesis of other g-butyrolactones (GBLs), such as the Streptomyces coelicolor butyrolactones and the Streptomyces virginiae butanolides (Figure 1d). [27,28] Members of this enzyme family are also involved in the biosynthesis of other classes of Streptomyces signalling molecules,s uch as 2-alkyl-4-hydroxymethylfuran-3-carboxylic acids (AHFCAs), typified by the methylenomycin furans, and 2-alkyl-4-hydroxy-3-methylbutenolides (AHMBs), exemplified by the Streptomyces rochei butenolides and the Streptomyces ansachromogenes butenolides (Figure 1d). [29][30][31] Incorporation experiments with stereospecifically 13 C-labelled glycerols indicate that an analogous phosphorylated butenolide to that formed by AfsA in A-factor biosynthesis is an intermediate in the biosynthesis of the methylenomycin furans. [32] This suggests that the AfsA family of enzymes involved in GBL, AHFCAand AHMB assembly all produce analogous phosphorylated butenolides that are diversified by subsequent biosynthetic enzymes ( Figure 1d).
PKSs typically have athioesterase (TE) domain appended to the C-terminus of the last module,which catalyses release of the fully assembled polyketide chain via hydrolysis or macrocyclisation, but several other chain release mechanisms are known. [33][34][35][36] Based on the role played by AfsA family enzymes in Streptomyces signalling molecule biosynthesis,we hypothesised that the AfsA-like domain appended to the Cterminus of the last module in the cryptic B. gladioli PKS releases the fully assembled polyketide chain by condensing it with DHAP to form ap hosphorylated butenolide (Figure 1d). Accordingly,w ed esignated this new type of chain release enzyme ap hosphorylated butenolide synthase (PBS) domain. Overall, our in silico analyses indicated that the cryptic B. gladioli PKS assembles an ovel glutarimidecontaining polyketide with significant structural differences to the glutarimide antibiotics assembled by Streptomyces species.

Isolation and Structure Elucidation of Metabolic Products of the Cryptic PKS
We originally identified the BGC encoding the cryptic trans-ATP KS in B. gladioli BCC0238. However, due to difficulties with creating in-frame deletions in this strain, we searched the genomes of other B. gladioli strains to see if they contain this BGC.A nother CF isolate, B. gladioli BCC1622, which is more amenable to genetic manipulation, [11] was also found to contain the cluster.U HPLC-ESI-Q-TOF-MS anal-ysis of an ethyl acetate extract from ac ulture of B. gladioli BCC1622, grown for 3days on am inimal agar medium containing glycerol as the sole carbon source,i dentified am etabolite with the molecular formula C 23 H 37 NO 5 (Figure S3).
Theplanar structure of this metabolite (5)was elucidated using 1 H, 13 Table S1). HMBC correlations between H-3 and C-1/C-1' and the exchangeable NH proton and C-2/C-2' confirmed the presence of a2 ,6 -piperidine- dione,a nd the chemical shift values for this moiety were in good agreement with the literature. [18,19,37,38] Twonetworks of COSY correlations established the structures of the C-2/C-2' to C-6 and C-8 to C-18 regions of the molecule,a nd HMBC correlations further confirmed the locations of the C-8 Me group and the C-9/C-10 double bond. HMBC correlations also showed that C-6 and C-8 are connected via aketo group and that amethyl ketone is attached to C-18. Based on a 3 J HH coupling constant of 15 Hz, the C-9/C-10 double bond was assigned the E configuration.
Although compound 5 contains the anticipated 2, 6piperidinedione,i tl acks the butenolide predicted to be installed by the PBS domain (Figure 1d). We thus postulated that this compound results from degradation of the true metabolic product of the BGC.Arange of carbon sources (glycerol, glucose,r ibose) and growth periods were explored to investigate whether other metabolic products of the BGC could be identified. UHPLC-ESI-Q-TOF-MS analysis of an ethyl acetate extract from a2 4hour culture on am inimal medium containing glucose as the sole carbon source identified an ew metabolite with the molecular formula C 27 H 39 NO 8 (Figure 2b and Figure S9). Atime course showed that production of this metabolite peaked at 20 hours and fell off rapidly over more protracted growth periods ( Figure S10). After 55 hours,s mall amounts of compound 5 could be detected, consistent with this being adegradation product of the new metabolite.
Comparison of the molecular formulae for compound 5 and the newly identified metabolite showed the latter contains four additional carbon atoms,a nd three additional hydrogen and oxygen atoms.T he planar structure of the new metabolite (6)w as elucidated using 1 H, 13 C, COSY,H SQC and HMBC NMR experiments (Figure 2a,F igure S11-S15, Table S2). The 13 CNMR spectrum of 6 lacked the signal due to methyl ketone in 5 and contained additional signals assigned to ac arbonyl group (C-21), two fully substituted alkene carbons (C-20 and C-2''), am ethyl group (C-3'')a nd ah emiacetal (C-1''). HMBC correlations between the C-3'' protons and C-1''/C-20 and the C-1'' proton and C-2''/C-21 led us to propose that this molecule contains a2 -subsituted 4hydroxy-3-methylbutenolide.T he NMR data for this moiety are similar to those reported for the AHMBs isolated from S. rochei and S. ansachromogenes (Figure 1d). [30,31] An HMBC correlation between the C-18 protons and C-20 established the connectivity between the butenolide and the rest of the structure.T he juxtaposition of a2 -acyl-4-hydroxy-3-methylbutenolide and 2, 6-piperidinedione in this molecule,coupled with the time course data for its production, led us to conclude that this compound, which we named gladiostatin, is the true metabolic product of the cryptic glutarimide-like BGC in B. gladioli BCC1622 and BCC0238. Gladiostatin (6)i sl ikely degraded to 5 via conjugate addition of water to the butenolide,f ollowed by ring opening,d ecarboxylation and retro-Aldol cleavage ( Figure S16).

Biological Activity of Gladiostatin
Initially,w et ested the activity of gladiostatin (6)a gainst representative members of the ESKAPE panel of bacterial pathogens, Candida albicans and Saccharomyces cerevisiae. While no activity was detected against the bacterial pathogens or C. albicans at concentrations up to 64 mgmL À1 ,gladiostatin was found to be active against S. cerevisiae,w ith an MIC of 4 mgmL À1 (Table 1).
Prompted by reports that several glutarimides have antitumour activity, [23,24] we investigated the activity of gladiostatin (6)against arange of cancer cell lines (Table 2). It was found to be active against ovarian, pancreatic and colon cancer cell lines ( Table 2). These values are in the same range as those reported for cycloheximide (1), migrastatin (3)a nd lactimidomycin (4)a gainst various other cell lines (Table  S6). [17,19,38] Interestingly,g ladiostatin (6)w as found to be inactive against the A549 lung cancer cell line,i ndicating it may exhibit some selectivity.
Some glutarimides have also been reported to inhibit tumour cell migration. [23,24] Thus,w eu sed aw ound-healing assay to test whether gladiostatin (6)can inhibit the migration of A2780 ovarian cancer cells.S trong suppression of cell migration was observed after 24 he xposure to 240 nM gladiostatin (Figure 3, Figure S17).

Proposed Pathway for Gladiostatin Biosynthesis.
To establish that the gds cluster directs the biosynthesis of gladiostatin (6), gdsE,w hich encodes the first polyketide synthase subunit, was disrupted in B. gladioli BCC1622 by insertional mutagenesis.LC-MS comparison of extracts from the wild type and mutant strains confirmed that gladiostatin (6)p roduction is abolished in the mutant (Figure 4).
Theearly stages of gladiostatin (6)biosynthesis appear to parallel the assembly of glutarimides in Streptomyces species. GdsB and GdsD are proposed to create aG dsC-bound malonamyl thioester starter unit, which is elaborated to a2 -  Figure 5). Consequently,gladiostatin (6)has amethyl branch at C-8, but not C-10. Additional studies are needed to understand how PKSs with seemingly identical architectures are able to produce the distinct C-methylation patterns observed in gladiostatin (6)a nd the Streptomyces glutarimides.
Thed omain architecture of modules 5-9 of the gladiostatin PKS diverges significantly from the corresponding modules in the Streptomyces glutarimide assembly lines. Indeed the C-terminus of module 4a ppears to be ab ranch point in all glutarimide-producing PKSs (Figure 1b), and could be ar ecombination hotspot. Based on comparisons between the domain architecture and the predicted biosynthetic intermediates,m odule 5o ft he gladiostatin PKS appears to lack both ak etoreductase (KR) domain (which is required for installation of the b-hydroxy group) and an enoyl reductase (ER) domain (required to saturate the double bond introduced by the dehydratase (DH) domain). Similarly,module 6lacks aDHdomain and modules 6, 7and 8 lack ER domains,s uggesting ah igh degree of non-linear programming in the gladiostatin PKS,w hich is ac ommon feature of glutarimide assembly lines (Figure 1). [39]     While it is unclear which domains are responsible for the keto reduction and dehydration reactions in modules 5and 6, respectively,i ns ilico analysis of the gladiostatin BGC identified two genes encoding putative reductases that could function as trans-acting ERs in modules 5-8. The gdsB gene encodes at ri-domain protein with acyl hydrolase (AH) and AT domains fused to af lavin-dependent ER domain and gdsH encodes an NAD(P)H-dependent oxidoreductase.W e therefore propose that one,o rb oth, of these reductases catalyse enoyl reduction in modules 5, 6, 7and 8ofthe PKS.A recently reported co-evolutionary categorisation of trans-AT PKS ACPd omains supports this hypothesis,i ndicating that the module 6, 7a nd 8A CP domains in the gladiostatin PKS clade with ACPd omains in modules with as imilar architecture from other assembly lines (i.e.K S-DH-KR-ACP and at rans-acting ER). [40] Moreover,t he TransATors oftware predicts that the KS domains in modules 7, 8a nd 9a re selective for an a,b-saturated thioester intermediate. [41] However,this software also predicts that the module 6KSdomain prefers a a,b-unsaturated thioester, highlighting potential limitations of such predictive bioinformatics analyses (Figure 5, Figure S18 and S19;T able S4 and S7).
Themost striking difference between the gladiostatin PKS and the other glutarimide assembly lines is the mechanism for polyketide chain release.All of the Streptomyces glutarimide PKSs are proposed to use TE domains that catalyse hydrolysis or macrolactonisation. In contrast, the gladiostatin PKS appears to employ aP BS domain to catalyse condensation of the b-ketothioester attached to the last module of the PKS with DHAP.T he resulting phosphorylated butenolide is proposed to undergo rearrangement and dephosphorylation to afford a2 -acyl-4-hydroxy-3-methylbutenolide ( Figure 5). This hypothesis is consistent with the observed increase in gladiostation production levels when glucose is used as the carbon source,b ecause DHAP is an intermediate in glycolysis.T oo ur knowledge,t here is no precedent for PKS chain release by aP BS domain. However,P BSs are known to catalyse the condensation of DHAP with ar ange of bketothioesters in the biosynthesis of several distinct classes of Streptomyces signalling molecule (Figure 1d). These include the SRBs and the SABs,w hich contain as imilar butenolide moiety to gladiostatin (6). In addition to the PBS (SabA), aphosphatase (SabP) and aketoreductase (SabD) are known to be required for SAB biosynthesis,b ut the mechanism for  Table S3. The stereochemistry of C-5 and C-8 is hypothesised to be the same as the corresponding stereocentres in lactimidomycin. Comparative sequencea nalysis of the module 2K Rdomain predicts that it produces an R-configured alcohol (Table S5), consistent with this hypothesis. elaboration of the putative phosphorylated butenolide produced by SabA to the 2-akyl-4-hydroxy-3-methylbutenolide remains to be elucidated. The gdsA and gdsG genes in the gladiostatin BGC encode putative phosphatases that could play as imilar role to SabP in the biosynthesis of the SABs (Table S3).

In vitro Reconstitution of Chain Release from the Gladiostatin PKS
To validate the proposed role of the PBS domain in gladiostatin (6)b iosynthesis,w ei nvestigated its catalytic activity using simplified mimics of the fully assembled polyketide chain. As ynthetic gene encoding the C-terminal ACPand PBS domains of GdsF was used to overproduce the ACP-PBS di-domain in E. coli as an N-terminal His 8 -fusion, which was purified to homogeneity using immobilised metalion affinity chromatography.T he identity of the purified protein was confirmed by ESI-Q-TOF-MS analysis (Figure S20).
Theability of the PBS domain to offload a3-ketothioester from the PKS was examined by loading a3-ketobutyryl mimic of the fully assembled polyketide chain onto the apo-ACP domain using the promiscuous phosphopantetheinyl transferase Sfp ( Figure S21). Themass of the protein decreased by 86 Da when DHAP was added ( Figure S21), consistent with cleavage of the 3-ketobutyryl group from the ACPd omain.
Incubation of the apo-ACP-PBS di-domain with the Nacetylcystetamine (NAC)t hioester of 3-ketooctanoate and DHAP,followed by treatment with shrimp alkaline phosphatase yielded aproduct, absent from the negative control, that gave rise to ions with m/z = 213.11 and m/z = 235.09 (corresponding to the [M+ +H] + and [M+ +Na] + ,respectively,ofthe 2acyl-3-hydroxymethyl butenolide (8)) in UHPLC-ESI-Q-TOF-MS analyses ( Figure 6). As ynthetic standard of 8 had the same retention time and MS/MS fragmentation pattern as the product of the enzymatic reaction ( Figure 6).

Conclusion
Thediscovery of gladiostatin (6)asthe product of acryptic trans-ATP KS assembly line in B. gladioli BCC0238 and BCC1622 further expands the already rich specialised metabolic repertoire of these and related CF isolates. [6,9,11] As the first glutarimide antibiotic to be isolated from aG ramnegative bacterium, gladiostatin differs significantly from other members of this family,a ll of which are produced by Streptomyces species.W hile some of these structural differences (such as the fully saturated C-11 to C-18 chain and the 2-acyl-4-hydroxy-3-methylbutenolide) are reflected by alterations in PKS architecture,o thers (e.g.t he lack of aC -10 methyl group) are not. Thed iscovery of gladiostatin thus offers agolden opportunity to develop abetter understanding of the role played by horizontal gene transfer in trans-AT PKS evolution, which could facilitate biosynthetic engineering approaches to polyketide structural diversification.
In vitro characterisation of the AfsA-like PBS domain appended to the C-terminus of the gladiostatin PKS shows that it releases 3-keto thioesters from the upstream ACP domain by condensing them with DHAP.T his constitutes an ew mechanism for polyketide chain release,w hich could prove to be av aluable addition to the synthetic biology toolbox. Thep hosphorylated butenolide product of the PBS domain is analogous to the intermediate in A-factor biosynthesis produced by AfsA. Thesame kind of intermediate has been shown to be formed by MmfL (another AfsA homologue) in the biosynthesis of the methylenomycin furans. [42] Thus,i ta ppears that phosphorylated 2-acyl-3-hydroxymethylbutenolide intermediates are involved in the biosynthesis of structurally diverse natural products,i ncluding GBLs, AHFCAs,A HMBs and gladiostatin. However,w ith the exception of GBLs,f urther work is needed to understand the mechanisms by which these intermediates get elaborated into the final metabolic products.