Insect‐Associated Bacteria Assemble the Antifungal Butenolide Gladiofungin by Non‐Canonical Polyketide Chain Termination

Abstract Genome mining of one of the protective symbionts (Burkholderia gladioli) of the invasive beetle Lagria villosa revealed a cryptic gene cluster that codes for the biosynthesis of a novel antifungal polyketide with a glutarimide pharmacophore. Targeted gene inactivation, metabolic profiling, and bioassays led to the discovery of the gladiofungins as previously‐overlooked components of the antimicrobial armory of the beetle symbiont, which are highly active against the entomopathogenic fungus Purpureocillium lilacinum. By mutational analyses, isotope labeling, and computational analyses of the modular polyketide synthase, we found that the rare butenolide moiety of gladiofungins derives from an unprecedented polyketide chain termination reaction involving a glycerol‐derived C3 building block. The key role of an A‐factor synthase (AfsA)‐like offloading domain was corroborated by CRISPR‐Cas‐mediated gene editing, which facilitated precise excision within a PKS domain.


Annotation of the gladiofungin biosynthesis gene cluster
AntiSMASH version 5.0 [1] and PKS/NRPS analysis [2] were used to analyze the genome sequence of B. gladioli HKI0739 ( Figure 1A). NCBI BLAST (database: swissprot) or HHpred [3] (databases: PDB_mmCIF70_27_Apr, Pfam-A_v32.0, NCBI_Conserved_Domains(CD)_v3.18) were used to identify unusual domains and closest homologues of described proteins (Table S2). Gene cluster comparison using EasyFig2.3 The sequence similarity of gene clusters encoding glutarimide-forming biosynthetic assembly lines is visualized using the tool Easyfig 2.3. Color code representing the sequence similarity values is given in the figure. GC base content is shown for the gladiofungin biosynthesis gene cluster ( Figure S1A-B).

Sequence similarity network (SSN) and genome neighborhood network (GNN)
Sequence similarity networks and genome neighborhood networks were created using the enzyme function initiative: enzyme similarity tool (EFI-EST) and genome neighborhood tool (EFI-GNT). [4] The amino acid sequences of GlaE or GlaD (module 1 and 2) were submitted as the queries to the EFI-EST online tool to create SSNs. Subsequently, the results were forwarded to EFI-GNT to create the corresponding the genome neighborhood. The SSNs were manually inspected for clusters in which glutarimide biosynthesis machineries are encoded ( Figure S1B).

Detailed analysis of gla PKS:
For a detailed analysis of the gla PKS, the deduced amino acid sequences of GlaABDE were submitted as query to the EFI-EST (enzyme function initiative: enzyme similarity tool) [4] to create SSNs (sequence similarity networks), and the results were forwarded to the genome neighborhood tool (EFI-GNT). The SSN generated with GlaE (AfsA domain) did not reveal any correlations to other biosynthetic gene clusters than those of B. gladioli. The SSN using the amino acid sequence of GlaD as query, however, revealed three significant neural clouds ( Figure S1B): 1) three homologous gene clusters from related Burkholderia strains; 2) a biosynthetic gene cluster (BGC) in Streptomyces amphibiosporus for lactimidomycin (6) [5] one in Streptomyces platensis for iso-migrastatin (7) assembly, [5] and 14 similar gene clusters in Streptomyces spp. or related strains; 3) the streptimidone (8a, 8b) BGC in Streptomyces himastatinicus, [6] the BGC for cycloheximide (9) and actiphenol (10) in Streptomyces sp. YIM 56141, [7] and 42 similar gene clusters in other Streptomyces spp. or related strains (Figure S1A-C). Notably, all of the known encoded compounds (6-10) share a glutarimide pharmacophore.   Figure 1; gray, hypothetical gene; yellow, genes coding for transporter; orange, regulators, light brown, amidotransferase; black, phosphatase; blue, PKS; purple, oxidoreductase; dark red, dehydrogenase; green, additional genes.

Knockout-mutant generation by kanR insertion into glaD, and glaG
Plasmids used in this section (see Table S4 and Figure S4) were designed to chromosomally integrate a kanamycin resistance marker (kanR) cassette into glaD (pJK345), or glaG (pJK347) via homologous recombination. Homologous flanking regions (F1 and F2) of the gene of interest were amplified by Phusion polymerase (New England Biolabs) from B. gladioli HKI0739 genomic DNA using the primers listed in Table S5. All flanking regions amplified were between 500 to 650 bp in size. The kanR cassettes were amplified from pGEM-Kan [8] by OneTaq polymerase (New England Biolabs) using the primers listed in Table S5. For the glaD and glaG knockout plasmids pJK345 and pJK347, XbaI/XhoI-digested pJET1.2 (Thermo Fisher Scientific) was assembled with the specific flanking regions F1 and F2 as well as the corresponding kanR cassette in a one-step reaction using NEBuilder HiFi DNA Assembly mix (New England Biolabs).
Following creation of the gene deletion plasmids, fresh B. gladioli HKI0739 cells growing on NAG agar were inoculated into 20 mL of MGY+M9 medium and incubated at 30 °C and with shaking at 110 rpm until cells reached OD 600 of 0.8-1.4. The cells were harvested by centrifugation at 6,000 × g at room temperature for 5 min and the precipitated cells were washed twice in 20 mL MiliQ water. The resulting cell pellet was resuspended in 500 µL MiliQ water. For each transformation, 100 µL of these cells were transformed with 0.5-1 µg of pJK345, pJK347, or pJK348 by electroporation. All electroporations were performed at 2.5 kV in a 2 mm gapped electroporation cuvette, followed by the addition of 500 µL of MGY+M9 medium. After incubation at 30 °C and with shaking at 100 rpm for 3 h, the cells were plated on NAG-kanamycin (300 µg mL -1 ) agar plate and incubated at 30 °C for 2 days.
Chromosomal integration of the kanamycin marker gene into glaD/G was verified by colony PCR using KAPA2G Robust HotStart ReadyMix PCR kit (Sigma-Aldrich). The primers were designed to detect the integration of the kanR cassette into ΔglaD (JK556/Kan-seq_rv), or ΔglaG (JK573/Kan-seq_rv), respectively. All positive colonies were counter screened to ensure they lacked any wildtype contamination and were not merely single crossover mutants (using the primers JK556/JK575 for pJK345, and JK560/JK574 for pJK347). Finally, positive colonies of each mutant were selected and the entire mutated gene region amplified for sequencing (Genewiz) and further analysis (using primer pairs JK556/JK557 for ΔglaD, and JK560/JK561 for ΔglaG). The obtained mutants are listed in Table S3. A scheme as well as the products of the final PCR for generation of the glaD mutant is shown in Figure S4.

Generation of CRISPR/Cas mutants containing a mutated or deleted AfsA domain
To establish a working CRISPR/Cas system for Burkholderia gladioli, two methods were combined: 1) A rhamnose inducible Burkholderia-optimized red-operon to enhance homologous recombination; [9] and 2) A single-copy, inducible, codon-optimized cas9 [10] to cause a doublestrand break at a specific position as determined by the N20-sequence of the sgRNA. All strains, plasmids, and primers used in this section are listed in Tables S3-5, respectively. The temperature-sensitive low-copy plasmid with a Burkholderia-optimized recombination system, pTsC-Red, was derived as follows. First, the temperature-sensitive pRO1600(Ts) replicon [11] and the genes coding for RedBA7029 [9] were constructed by gene synthesis (Life Technologies, ThermoFisher). Plasmid pMK-TsOriRO1600 was then BspHI/FspI-digested and ligated with a BspHI-digested chloramphenicol resistance marker (cmR) cassette amplified from pCR-cmR (a subcloning plasmid that contains the cmR cassette from pACYCDuet-1 from Novagen) using the primers JK435/JK436 to yield plasmid pMC-TsOriRO1600. Next, redγ was amplified from pKD46-Gm [12] using the primers JK397/JK398 and redβα7029 was amplified from pMK-RedBA7029 using the primers JK399/JK400. Both PCR products were assembled into NdeI/SbfI-digested pSCrhaB2 to give pRedBA7029. Finally, the rhaSR-P rhaB -redγβα7029 cassette was cut from pRedBA7029 using BsrGI/SbfI and ligated in between the BsrGI/SbfI sites of pMC-TsOriRO1600 to create pTsC-Red, a helper plasmid that can be used for homologous recombination in diverse Burkholderia strains.
To create pTsK-CasRed-Bt, a one-step CRISPR/Cas plasmid based on pTsC-Red, a codonoptimized, Streptococcus pyogenes derived cas9* gene in combination with the rhaB promoter was synthesized (Genewiz). The codon-optimization analysis was performed by the GENEius web tool (http://www.geneius.de) using the codon-usage table of Burkholderia thailandensis. The cmR cassette from pTsC-Red was replaced by a kanR cassette by digesting pTsC-Red using PspOMI/BspEI and inserting the kanR cassette (amplified from pGEM-Kan [8] using the primers JK678/679) using NEBuilder isothermic assembly, producing pTsK-Red. Next, the P rhaB -cas9*cassette was cut from its subcloning vector by digestion with PspOMI/SbfI and ligated into pTsK-Red, cut with the same restriction enzymes, to yield pTsK-CasRed-Bt. Cassettes containing the specific guide RNA (sgRNA) for attack at the N20-PAM sequence CAT CGA CTG ACG GGC AGC CT-CGG as well as the alternative DNA for genome editing were obtained by gene synthesis (Genewiz) and are displayed in Table S6. These cassettes were excised from their subcloning plasmids using PspOMI/XhoI and ligated into pTsK-CasRed-Bt, yielding plasmids pJK363 for afsA point mutation (PM) and pJK364 for afsA-domain deletion, respectively ( Figure S5). Both plasmids were introduced into B. gladioli HKI0739 cells by electroporation as described above. Cells were plated on NAG-kanamycin (300 µg mL -1 ) agar supplemented with 2.5 mg mL -1 L-rhamnose (NAG-Kan-Rha) and incubated at 30 °C for 2 days. Up to 30 kanamycin-resistant colonies were picked on a fresh NAG-Kan-Rha plate and again incubated at 30 °C for 2 days. This was repeated twice until colonies displaying reduced growth were detected. These colonies were checked by colony-PCR using primers JK725/JK610 (protocol A). The afsA domain coding-frame deletion mutant was detected by the expected product size (1,401 bp instead of 1,971 bp for the wild type). To identify the afsA-PM mutant (E10080A), the PCR product was digested with SacII. Clones containing the point mutation showed a fragment pattern of 837 bp + 354 bp + 406 bp + 374 bp instead of the wild-type pattern of 1,191 bp + 406 bp + 374 bp. To verify that wild-type contamination was absent, all positive colonies were streaked out on NAG-Kan-Rha plate again, and single colonies were checked using the same method as before. In addition, a second PCR using primers JK725/JK728 (protocol B) was performed that only gave a product (856 bp) if contaminating wild type cells were present. To cure the plasmid, 1-2 positive colonies were streaked on a NAG agar plate (without kanamycin and rhamnose) and incubated at 37 °C for 16 hours. Grown colonies were tested for kanamycin sensitivity, indicating plasmid loss. Again, positive colonies were screened using colony PCR (protocols A and B) to ensure the correct genotype and the absence of wild type cells. For final verification, the product of PCR protocol A was sequenced (Genewiz) ( Figure S6 and Figure S7).

Off-target analysis in the B. gladioli afsA-del mutant
The three most prpbable gene regions for off-target effects were chosen by a BLAST search in combination with the presence of a PAM sequence (NGG) ( Table S7). They were amplified and sequenced (Genewiz) from the B. gladioli WT and the afsA-deletion mutant using the primers listed in Table S5 and the KAPA2G Robust HotStart ReadyMix PCR kit (Sigma-Aldrich). Sequence alignment showed no discrepancy in any of the targets ( Figure S8) indicating that there is only little risk for unwanted gene editing elsewhere in the genome. In addition, no significant difference in growth behavior could be observed ( Figure S10).
Furthermore, the B. gladioli afsA-PM mutant that was generated using the identical sgRNA as for the afsA-del mutant shows a comparable metabolic profile as the WT strain ( Figure S9). This indicates that at least none of the BGC of detectable metabolites was significantly altered.

Transcriptome analysis of the B. gladioli afsA-del mutant
To determine strong polar effects of the CRISPR/Cas generated gene region deletion on the transcription of the adjacent downstream genes (glaF and glaG), a qualitative RT-PCR was performed. Therefore, the B. gladioli WT and the B. gladioli afsA-del mutant were grown in PDB for 32 h at 30 °C, 120 rpm ( Figure S10). Samples were taken during late exponential (OD 600 of 0.8-1.0; 10 mL), stationary (after 8 h; OD 600 ≈2.5; 4 mL), and late stationary growth phase (after 32 h; OD 600 ≈10; 1.5 mL). For RNA extraction the Quick RNA Fungal Bacterial Miniprep Kit (Zymo Research) was used in combination with the TURBO DNA-free kit (ThermoFisher). After quantification via nanodrop approx. 2.5 µg of total RNA were treated in a 20 µL reaction using random primers and the Thermo Maxima H Minus First Strand cDNA Synthesis Kit (ThermoFisher). A non-template control (NTC) as well as a negative control for each sample (-RT) were included.
For PCR 1 µL template was added in a 10 µL reaction using the KAPA2G Robust HotStart ReadyMix PCR kit (Sigma-Aldrich). All samples were analyzed with primers JK560 and JK574 to determine the presence of specific glaF and glaG cDNA (PCR A). In addition, samples from stationary growth phase were also tested with primers JK556/JK575 for glaD transcription (PCR B) and with primers JK725/JK728 for the presence of afsA specific cDNA (PCR C), respectively.For further verification, the product of PCR A was sequenced (Genewiz) As shown in Figure S11, transcription of glaF and glaG could be proven in the WT strain as well as in the afsA-del mutant during all examined growth phases. As expected, both strains also did not differentiate regarding glaD transcription, but concerning the presence of the mRNA afsA-region.
These results in combination with the present gladiofungin production in the B. gladioli afsA-PM mutant ( Figure S9) strongly support the hypothesis that the CRISPR/Cas genome editing did not lead to polar effects in the B. gladioli mutants.

Phylogenetic analysis of ketosynthase domains of gladiofungin PKS
A selection of amino acid sequences of KS domains from trans-AT assembly lines were used to deduce the KS specificities. [15] The amino acid sequences of the gladiofungin KS sequences were assigned according to antiSMASH analyses. [16] The multiple sequence alignment was performed using MEGA6 [17] with default settings. A molecular phylogenetic tree was constructed by the Maximum Likelihood method by using IQ-tree web server (Ultrafast bootstrapping, 1000 iterations). [18] The analysis involved 204 KS amino acid sequences ( Figure S12). The ketoreductase (KR) specificity of the domain in module 3 of the gladiofungin assembly line was predicted based on the method by Caffrey, [19] correlating the conserved HXXXXXXD motif with a D-configuration of the β-hydroxy group. KR domain amino acid sequences were aligned using Mega6 [17] and a Maximum Likelihood Tree was constructed by IQ-tree web server. [18] Based on the diagnostic amino acid sequence (HAAGTLRD), the configuration at C-7 was predicted to be R.
The stereochemistry of C-10 attached to the methyl group in 6 was determined by using the amino acid sequence of the associated module 5 KR according to previous studies [20] (https://akitsche.shinyapps.io/profileHMM_App/). The methyl group was predicted with a ScoreDiff value of -108.83 (and an a priori known D-form of the secondary alcohol) as the L-form.

Determination of bioactivity
Antibacterial and antifungal profiling was carried out in an agar diffusion assays as previously described (Table S8). [21] In addition, gladiofungin A (dissolved in methanol) was tested in different concentrations in an agar diffusion assay against the entomopathogenic Purpureocillium lilacinum ( Figure S13 and Table S9). Cytotoxic and antiproliferative activities were tested as described previously (Table S10). [22] Here, all substances were dissolved in DMSO.       Detailed structure elucidation The structure of 1 was elucidated by 1D-and 2D-NMR. 13 C NMR and DEPT-135 data of 1 indicated the presence of 6 methines, 12 methylenes, 2 methyl groups, and 7 quaternary carbons, from which 5 were assigned to carbonyls. In extensive 1 H-1 H COSY analyses, three spin systems (from H-2 to H-8, from H-10 to H-16, and H-19 to H-21) were observed. The first fragment (H-2 to H-8) contains the glutarimide moiety, judging from the 1 H-13 C HMBC correlations from NH (δ 8.64 ppm)/H-2 to C-1 (δ 172.1 ppm) and from NH/H4-to C-5 (δ 172.2 ppm). The second fragment (H-10 to H-16) was connected via keto carbonyl carbon C-9 (δ 212.7 ppm) with the first fragment by the HMBC correlations between H-8/H-10 and C-9. The HMBC correlation from H-21 to the keto carbonyl carbon C-22 (δ 199.2 ppm) connects with the third fragment (C-19 to H-21). A quaternary carbon rich heterocyclic moiety, which is adjacent to C-22, was deduced as trisubstituted butenolide by four HMBC correlations from singlet methyl protons 27-CH 3 (δ 2.24 ppm) to C-22/C-23 (δ 128.0 ppm)/C-25 (δ 98.9 ppm) and from singlet hemiacetal methine proton H-25 (δ 5.89 ppm) to C-24 (δ 169.7 ppm). To increase 13 C NMR signals in this heterocyclic moiety, 1 was 13 C-enriched by isotope labeling using ubi-13 C 3 -glycerol. The ADEQUATE correlations from 27-CH 3 /H-25 to C-26 and the INADEQUATE correlation between C-23 and C-24, respectively, obtained from 13 C-enriched 1 supported to elucidate the five-membered ring structure ( Figure 3B). Finally, the observed INADEQUATE correlations from C16 to C19 completed the structure of 1.
The E-configuration of the double bond C-12/C-13 (δ H 5.36 and 5.60 ppm) was assigned based on the proton coupling constant J 12-13 15.5 Hz.