Resistance-Gene Directed Discovery of a Natural Product Herbicide with a New Mode of Action

Abstract

Text Bioactive natural products have evolved to inhibit specific cellular targets and have served as lead molecules for health and agricultural applications for the last century 1–3 . The post-genomics era has brought a renaissance in natural product discovery using synthetic biology tools 4–6 . However, compared to traditional bioactivity-guided approaches, genome mining of natural products with specific and potent biological activities remains challenging 4 . Here we present the discovery and validation of a highly potent herbicide lead that targets a critical metabolic enzyme that is required for plant survival. Our approach is based on the coclustering of a self-resistance gene in the natural product biosynthetic gene cluster 7–9 , which serves as a window to potential biological activity of the encoded compound. We targeted dihydroxyacid dehydratase (DHAD) in the branched-chain amino acid (BCAA) biosynthetic pathway in plants, the last step in this pathway often targeted for herbicide development 10 . We showed the fungal sesquiterpenoid aspterric acid (AA) discovered using this method is a submicromolar inhibitor of DHAD and is effective as an herbicide in spray applications. The self-resistance gene astD was validated to be insensitive to AA and was deployed as a transgene in establishment of plants that are resistant to AA. This herbicide-resistance gene combination complements urgent efforts in overcoming weed resistance 11 . Our discovery demonstrates the potential of using a resistance-gene directed approach in mining of bioactive natural products. Weeds are a major source of crop losses, and the evolution of herbicide resistance in weeds has led to an urgent need for new herbicides with novel modes of action 11–14 . The BCAAs biosynthetic pathway is essential for plant growth 10 . It is not present in animals and is therefore a validated target for highly specific weed control agents 10 . The BCAA biosynthetic pathway in plants is carried out by three enzymes: acetolactate synthase (ALS), acetohydroxy acid isomeroreductase (KARI), and DHAD (Fig. 1a). Given the success of targeting ALS for herbicide development 11 , it is surprising that no herbicide that targets either of the other two enzymes has been developed. DHAD which catalyzes β-dehydration reactions to yield α-keto acid precursors to isoleucine, valine and leucine, is an essential and highly conserved enzyme among plant species 15,16 (Extended Data Fig. 1a and Supplementary Fig. 1). Efforts toward synthetic DHAD inhibitors resulted in compounds with submicromolar K i; however, the compounds have no in planta activity 17 (Extended Data Fig. 1b). Filamentous fungi are prolific producers of natural products (NPs), many of which have biological activities that aid the fungi in colonizing and killing plants 1,2,18 . Therefore, fungal NPs represent a promising source of potential leads for herbicides. The abundance of sequenced fungal genomes enables genome mining of new NPs with novel biological activities 4,6 . Although no NP inhibitors of DHAD are known to date, we reason a fungal NP with this property might exist, given the indispensable role of BCAA biosynthesis in plants 10 . To identify NP biosynthetic gene clusters that may encode a DHAD inhibitor, we hypothesized that such cluster must contain an additional copy of DHAD that is insensitive to the inhibitor, thereby providing the required self-resistance for the producing organism to survive. The presence of a gene encoding a self-resistance enzyme is frequently found in microbial NP gene clusters, as highlighted by the presence of an insensitive copy of HMGR or IMPDH in the gene clusters for lovastatin (that targets HMGR) or mycophenolic acid (that targets IMPDH), respectively (Extended Data Fig. 1c) 19,20 . This phenomenon has been used to predict molecular targets of NPs, as well as to identify gene clusters of NPs of known activities 5,7,9 . To identify possible self-resistance enzymes, we scanned sequenced fungal genomes to search for colocalizations of genes encoding DHAD with core biosynthetic enzymes, such as terpene cyclases, polyketide synthases, etc 21,22 . We identified a well-conserved set of four genes across multiple fungal genomes (Fig. 1b), including the common soil fungus Aspergillus terreus that is best known to produce lovastatin. The conserved gene clusters include genes that encode a sesquiterpene cyclase homolog (astA), two cytochrome P450s (astB and astC), and a homolog of DHAD (astD). Genes outside of this cluster are not conserved across the identified genomes and are hence unlikely to be involved in NP biosynthesis. AstD is the second copy of DHAD encoded in the genome, and is ~70% similar to the housekeeping copy that is well-conserved across fungi (Supplementary Fig. 2). Therefore, AstD is potentially a self-resistance enzyme that confers resistance to the encoded NP. Like a majority of biosynthetic gene clusters in sequenced fungal genomes, the ast cluster has not been associated with the production of a known NP 4 . To identify the NP encoded by the ast cluster, we heterologously expressed astA, astB, and astC genes in the host Saccharomyces cerevisiae RC01 23 . New compounds that emerged were purified and their structures were elucidated with NMR spectroscopy (Supplementary Fig. 3 and Supplementary Table 5). RC01 expressing only astA produced a new sesquiterpene (1), which was confirmed to be (−)-daucane (Supplementary Fig. 4). RC01 expressing both astA and astB led to the biosynthesis of a new product that was structurally determined to be the α-epoxy carboxylate (2) (Fig. 1c). When astA, astB and astC were expressed together, a new compound (3) became the dominant product (~ 20 mg/L). Full structural determination revealed the compound to be the tricyclic aspterric acid (AA), which is a previously isolated compound (Fig. 1c) 24 . The biosynthetic pathway for AA is therefore concise: following cyclization of farnesyl diphosphate by AstA to create the carbon skeleton in 1, AstB catalyzes oxidation of 1 to yield the epoxide 2. Further oxidation by AstC at carbon 15 yields an alcohol, which can undergo intramolecular epoxide opening to create AA (Fig. 1d). Upon its initial discovery, AA was shown to have inhibitory activity towards Arabidopsis thaliana, however, the mode of action was not known 25 . Our resistance-gene directed approach led to rediscovery of this compound with DHAD as a potential target. We first confirmed that AA is able to potently inhibit A. thaliana growth in an agar-based assay (Fig. 2a, and Supplementary Fig. 5). AA was also an effective inhibitor of root development and plant growth when applied to a representative monocot (Zea mays) and dicot (Solanum lycopersicum) (Fig. 2b). To test if AA indeed targets DHAD, we expressed and purified housekeeping DHAD from both A. terreus (XP_001208445.1, fDHAD) and A. thaliana (AT3G23940, pDHAD), as well as the putative self-resistance enzyme AstD (Supplementary Fig. 6). Both housekeeping DHAD enzymes converted dihydroxyisovalerate to ketoisovalerate (pDHAD: k cat = 1.2 sec−1, K M = 5.7 mM) as expected. The enzyme activities, however, were inhibited in the presence of AA (Extended Data Fig. 2). The IC50 values of AA towards fDHAD and pDHAD were 0.31 μM and 0.50 μM at an enzyme concentration of 0.50 μM, respectively (Extended Data Fig. 3). AA was further determined to be a competitive inhibitor of pDHAD with a K i = 0.30 μM (Extended Data Fig. 3). AA displayed no significant cytotoxicity towards human cell lines up to 500 μM concentration, consistent with the lack of DHAD in mammalian cells (Supplementary Fig. 7). AstD catalyzes the identical β-dehydration reaction as DHAD, albeit with a significantly more sluggish turnover rate (k cat = 0.03 sec−1, K M = 5.4 mM). However, the enzyme was not inhibited by AA, even at the solubility limit of 8 mM (Extended Data Fig. 3). To determine if AstD can confer resistance to AA-sensitive strains, we developed a yeast based assay. The genome copy of DHAD encoded by ILV3 was first deleted from Saccharomyces cerevisiae strain DHY ΔURA3, which resulted in an auxotroph that requires exogenous addition of Ile, Leu and Val to grow. We then introduced either fDHAD or astD episomally, both of which allowed the strain to grow in the absence of the three BCAAs (Extended Data Fig. 4). However, yeast expressing fDHAD was approximately 100 times more sensitive to AA (IC50 of 2 μM) compared to yeast expressing AstD (IC50 of 200 μM) (Fig. 2c). Collectively, the biochemical and genetic assays validated AA as the first natural product inhibitor of fungal and plant DHAD; and AstD serves as the self-resistance enzyme in the ast biosynthetic gene cluster. The (R)-α-hydroxyacid and (R)-configured β-ether oxygen moieties in AA mimic the (2R, 3R)-dihydroxy groups present in natural substrates such as dihydroxyisovalerate. The β-ether oxygen in AA is in position to coordinate to the 2Fe-2S cluster that is a required cofactor in both fungal and plant DHAD 16,17 . To understand potential AA mechanism of action, we determined the crystal structure (2.11 Å) of the pDHAD complexed with 2Fe-2S cluster (holo-pDHAD) (Fig. 2d, Extended Data Fig. 5, and Extended Data Table 1). We identified a binding chamber at the homodimer interface, similar to that found in the holo bacterial L-arabinonate dehydratase 26 (Fig. 2d). The interior of the chamber is positively charged (2Fe-2S and Mg2+) while the entrance is lined with hydrophobic residues. The modeled binding mode of α,β-dihydroxyisovalerate and AA predicted by computational docking are shown in Fig. 2e. The pocket is sufficiently spacious to accommodate the bulkier AA, and provide stronger hydrophobic interactions than the native substrate with a 5.3 ± 0.3 kcal/mol gain in binding energy (Fig. 2e). Base on the holo-pDHAD structure, we constructed a homology model of AstD to determine potential mechanism of resistance (Extended Data Fig. 5 and 6). Comparison of pDHAD and the modeled AstD structures shows that while most of the residues in the catalytic chamber are conserved, the hydrophobic region at the entrance to the reactive chamber in AstD is more constricted as a result of two amino acid substitutions (V496L and I177L). Narrowing of the entrance could therefore sterically exclude the bulkier AA from binding in the active site, while the smaller, natural substrates are still able to enter the chamber. To explore the potential of AA as an herbicide, we performed spray treatment of A. thaliana with AA. We added AA into a commercial glufosinate formulation known as Finale® at a final AA concentration of 250 μM 27,28 . We then sprayed AA solution onto glufosinate resistant A. thaliana. Finale® alone had no observable inhibitory effects on plant growth, but adding AA severely inhibited plant growth (Extended Data Fig. 7). In addition, A. thaliana plants treated with AA before flowering failed to form normal pollen, which was also observed previously 25 . We found that the pistil of treated plants could still be successfully pollinated using healthy pollen from the untreated A. thaliana, indicating that AA preferentially affects pollen but not egg formation (Extended Data Fig. 8 and 9). This effect was also observed with a lower concentration of AA (100 μM). Thus, in addition to its herbicidal properties, AA could potentially be used as a chemical hybridization agent for hybrid seed production 29 . We next investigated whether plants expressing astD are resistant to AA. This was motivated by the successful combination of glyphosate and genetically modified crops that are selectively resistant to glyphosate (Roundup Ready®) 30 . The A. terreus astD gene was codon optimized and the N-terminus was fused to a chloroplast localization signal derived from pDHAD. Wild type or astD transgene-expressing A. thaliana was then grown on media that contained 100 μM AA. In the presence of AA, the growth of wild-type plants was strongly inhibited, and arrested at the cotyledon stage (Fig. 3a). In contrast, the growth of astD transgenic plants was relatively unaffected by AA, as indicated by the normally expanded rosette leaves, elongated roots, and whole plant fresh weight (Fig. 3a and b). The expression of AstD was verified by western blot (Supplementary Fig. 8). A spray assay was also performed using T2 astD transgenic A. thaliana plants, which showed no observable growth defects under such treatment (Fig. 3c). In contrast, the control plants carrying the empty vector showed a strong growth inhibitory phenotype when treated with AA (Fig. 3c). Quantitative measurements of plant height showed AstD effectively confers AA resistance to A. thaliana (Fig. 3d). In summary, resistance-gene directed discovery of NPs in the fungus A. terreus led to the discovery of a natural herbicide AA and the determination of its mode of action. In addition, introducing astD as a transgene or editing the sequence of the plant DHAD endogenous gene could be used to create AA-resistant crops. We suggest that AA is a promising lead for the development as a broad spectrum commercial herbicide. METHODS General materials and methods Biological reagents, chemicals, media and enzymes were purchased from standard commercial sources unless stated. Plant, fungal, yeast and bacterial strains, plasmids and primers used in this study are summarized in Supplementary Tables 3, 4 and 5. DNA and RNA manipulations were carried out using Zymo ZR Fungal/Bacterial DNA Microprep™ kit and Invitrogen Ribopure™ kit respectively. DNA sequencing was performed at Laragen, Inc. The primers and codon optimized gblocks were synthesized by IDT, Inc. Expression of ast genes in Aspergillus nidulans for cDNA isolation Plasmids pYTU, pYTP, pYTR digested with PacI and SwaI were used as vectors to insert genes 31 . A gpda promoter was generated by PCR amplification using primers Gpda-pYTU-F and Gpda-R with pYTR serving as template. Genes to be expressed were amplified through PCR using the genomic DNA of Aspergillus terreus NIH2624 as a template. A 4.5 kb fragment obtained using primers AstD-pYTU-recomb-F and AstA-pYTU-recomb-R was cloned into pYTU together with a gpda promoter by yeast homologous recombination to obtain pAstD+AstA-pYTU. Yeast transformation was performed using Frozen-EZ Yeast Transformation II Kit™ (Zymo Research). A 2.4 kb fragment obtained using primers AstB-pYTR-recomb-F and AstB-pYTR-recomb-R was cloned into pYTR by yeast homologous recombination to obtain pAstB-pYTR. Similarly, a 2.3 kb fragment obtained using primers AstC-pYTP-recomb-F and AstC-pYTP-recomb-R was cloned into pYTP by yeast homologous recombination to obtain pAstC-pYTP. All three plasmids (pAstD+AstA-pYTU, pAstB-pYTR and pAstC-pYTP) were transformed into A. nidulans following standard protocols to result in the A. nidulans strain TY01 31 . TY01 was cultured in liquid CD-ST medium (20 g/L starch, 20 g/L peptone, 50 mL/L nitrate salts and 1 mL/L trace elements) at 28°C for 3 days. Total RNA of TY01 was extracted with the Invitrogen Ribopure™ kit, and total cDNA of TY01 was obtained using the SuperScript III reverse transcriptase kit (Thermo Fisher Scientific). The cDNA fragment of astA was PCR amplified using primers AstA-xw55-recomb-F and AstA-xw55-recomb-R. The cDNA fragment of astB was PCR amplified using primers AstB-xw06-recomb-F and AstB-xw06-recomb-R. The cDNA fragment of astC was PCR amplified using primers AstC-xw02-recomb-F and AstC-xw02-recomb-R. The cDNA fragment of astD was PCR amplified using primers AstD-pXP318-F and AstD-pXP318-R. All the introns were confirmed to be correctly removed by sequencing. Construction of Saccharomyces cerevisiae strains Plasmid pXW55 (URA3 marker) digested with NdeI and PmeI was used to introduce the astA gene 23 . A 1.3 kb fragment containing astA obtained from PCR using primers AstA-xw55-recomb-F and AstA-xw55-recomb-R was cloned into pXW55 using yeast homologous recombination to afford pAstA-xw55. The plasmid pAstA-xw55 was then transformed into Saccharomyces cerevisiae RC01 to generate strain TY02 23 . Plasmid pXW06 (TRP1 marker) digested with NdeI and PmeI was used to introduce the astB gene 23 . A 1.6 kb fragment containing astB obtained from PCR using primers AstB-xw06-recomb-F and AstB-xw06-recomb-R were cloned into pXW06 using yeast homologous recombination to afford pAstB-xw06. The plasmid pAstB-xw06 was then transformed into TY02 to generate strain TY03. Plasmid pXW06 (LEU2 marker) digested with NdeI and PmeI was used to introduce the astC gene 23 . A 1.6 kb fragment containing astC obtained from PCR using primers AstC-xw02-recomb-F and AstC-xw02-recomb-R were cloned into pXW02 using yeast homologous recombination to afford pAstC-xw02. The plasmid pAstC-xw02 was then transformed into TY03 to generate strain TY04. URA3 gene was inserted into ilv3 locus of Saccharomyces cerevisiae DHY ΔURA3 strain to generate UB01. A 879 bp homologous recombination donor fragment with 35–40 bp homologous regions flanking ilv3 ORF was amplified using primers ILV3p-URA3-F and ILV3t-URA3-R using yeast gDNA as template. The PCR product was gel purified and transformed into Saccharomyces cerevisiae DHY ΔURA3, and selected on uracil dropout media to give UB01. The resulting strain was subjected to verification by colony PCR with primers ILV3KO-ck-F and ILV3KO-ck-R and the amplified fragment was sequence confirmed. The URA3 gene inserted into ilv3 locus of Saccharomyces cerevisiae DHY ΔURA3 was deleted from UB01 using homologous recombination to generate UB02. A 150 bp homologous recombination donor fragment with 75 bp homologous regions flanking ilv3 ORF was amplified using primers ILV3KO-F and ILV3KO-R, gel purified and transformed into UB01, and counterselected on 5-fluoroorotic acid (5-FoA) containing media to give UB02. The resulting strain was subjected to verification by colony PCR with primers ILV3KO-ck-F and ILV3KO-ck-R and the amplified fragment was sequenced confirmed. The empty plasmid pXP318 (URA3 marker) was transformed into UB02 to generate TY05 32 . Plasmid pXP318 digested with SpeI and XhoI was used as vector to introduce gene encoding fDHAD 32 . The cDNA of Aspergillus terreus NIH 2624 served as template for PCR amplification. A 1.7 kb fragment obtained using primers fDHAD-pXP318-F and fDHAD-pXP318-R were cloned into pXP318 using yeast homologous recombination to afford fDHAD-pXP318. Then, fDHAD-pXP318 was transformed into UB02 to generate TY06. fDHAD was driven by a constitutive promoter TEF1. Plasmid pXP318 digested with SpeI and XhoI was used as vector to introduce astD gene 32 . The cDNA isolated from TY01 served as the template for PCR amplification. A 1.8 kb fragment obtained using primers AstD-pXP318-F and AstD-pXP318-R were cloned into pXP318 using yeast homologous recombination to give AstD-pXP318. A FLAG-tag was also add to the N-terminal of AstD. Then, AstD-pXP318 was transformed into UB02 to generate TY07. AstD was driven by a constitutive promoter TEF1. Fermentation and compound analyses and isolation A seed culture of S. cerevisiae strain was grown in 40 mL of synthetic dropout medium for 2 d at 28°C, 250 rpm. Fermentation of the yeast was carried out using YPD (yeast extract 10 g/L, peptone 20 g/L) supplement with 2% dextrose for 3 d at 28°C, 250 rpm. HPLC-MS analyses were performed using a Shimadzu 2020 EVLC-MS (Phenomenex® Luna, 5μ, 2.0 × 100 mm, C-18 column) using positive and negative mode electrospray ionization. The elution method was a linear gradient of 5–95% (v/v) acetonitrile/water in 15 min, followed by 95% (v/v) acetonitrile/water for 3 min with a flow rate of 0.3 mL/min. The HPLC buffers were supplemented with 0.05% formic acid (v/v). HPLC purifications were performed using a Shimadzu Prominence HPLC (Phenomenex® Kinetex, 5μ, 10.0 × 250 mm, C-18 column). The elution method was a linear gradient of 65–100% (v/v) acetonitrile/water in 25 min, with a flow rate of 2.5 mL/min. GC-MS analyses were performed using Agilent Technologies GC-MS 6890/5973 equipped with a DB-FFAP column. An inlet temperature of 240°C and constant pressure of 4.2 psi were used. The oven temperature was initially at 60°C and then ramped at 10°C/min for 20 min, followed by a hold at 240°C for 5 min. To isolate compound 1, the fermentation broth of TY02 was centrifuged (5000 rpm, 10 min), and cell pellet was harvested and soaked in acetone. The organic phase was dried over sodium sulfate, concentrated to oil form, and subjected to silica column purification with hexane. To isolate compound 2, the fermentation broth of TY03 was centrifuged (5000 rpm, 10 min), and supernatant was extracted three times with ethyl acetate. The organic phase was dried over sodium sulfate, concentrated to oil form, and then and subjected to HPLC purification. To isolate compound AA, the fermentation broth of TY04 was centrifuged (5000 rpm, 10 min), and supernatant was extracted three times with ethyl acetate. The organic phase was dried over sodium sulfate, concentrated to oil form, and subjected to HPLC purification. Structure determination of compounds Compound 1, colorless oil readily dissolved in hexane and chloroform, had a molecular formula C15H24, as deduced from EI-MS [M]+ m/z 204, and showed [ α ] D 22 = - 30 ° (n-hexane; c = 0.1). GC-MS 70 eV, m/z (relative intensity): 204 [M]+ (42), 189 (5), 161 (35), 136 (100), 133 (10), 121 (70), 119 (25), 107 (20), 105 (27), 93 (21), 91 (26), 79 (13), 77 (15), 69 (20), 55 (12), 43 (12), 41 (13), 38 (21); 1H NMR (500 MHz, CDCl3): δ 5.37 (1H, m), 2.20-2.10 (5H, m), 2.10-2.00 (2H, m), 1.95 (1H, d, 15.3), 1.75 (3H, s), 1.71 (3H, q, 1.7), 1.61 (3H, brs), 1.44 (1H, dd, 11.4, 7.2), 1.36 (1H, m), 1.31 (1H, dd, 11.3, 2.6), 0.73 (3H, s); 13C NMR (125 MHz, CDCl3): δ 138.4, 138.3, 122.4, 122.2, 57.4, 42.6, 41.4, 40.3, 34.5, 29.6, 27.3, 25.0, 23.3, 20.6, 19.2. Both of the NMR and MS spectrums are identical to a known compound (+)-daucane, however, the optical rotation is opposite which led to the assignment of 1 to be (−)-daucane 33 . Compound 2, colorless oil readily dissolved in ethyl acetate and chloroform, had a molecular formula C15H22O3, as deduced from LC-MS [M+H]+ m/z 251, [M-H]− m/z 249. 1H NMR (500 MHz, CDCl3): δ 8.09 (1H, brs), 3.25 (1H, t, 7.4), 2.71 (1H, dd, 14.6, 6.5), 2.48 (1H, dd, 14.8, 6.3), 2.36 (1H, dd, 14.0, 6.6), 2.26 (1H, m), 2.15 (1H, dd, 16.3, 8.9), 2.08 (1H, d, 12.0), 1.84 (1H, q, 13.1), 1.73 (3H, d, 2.3), 1.59 (3H, d, 2.2), 1.48~1.35 (3H, m), 1.31 (1H, td, 11.5, 9.0), 0.86 (3H, s). 13C NMR (125 MHz, CDCl3): δ 176.0, 135.8, 123.2, 60.1, 59.8, 59.4, 44.1, 40.5, 38.8, 30.6, 29.3, 24.9, 23.8, 20.6, 17.8. Compound 3 is a colorless oil readily dissolved in acetone and chloroform, had a molecular formula C15H22O4, as deduced from LC-MS [M+H]+ m/z 267, [M-H]− m/z 265. 1H NMR (500 MHz, CDCl3): δ 4.29 (1H, d, 8.5), 3.92 (1H, d, 8.3), 3.48 (1H, d, 8.3), 2.42 (1H, dd, 14.9, 7.3), 2.37~2.28 (2H, m), 2.25 (1H, dd, 13.0, 4.4), 2.20~2.17 (1H, m), 2.12 (1H, d, 13.4), 2.01 (1H, m), 1.80~1.65 (2H, m), 1.71 (3H, s), 1.64~1.54 (1H,m), 1.60 (3H, s), 1.50 (1H, m); 13C NMR (125 MHz, CDCl3): δ 178.2, 134.5, 125.2, 82.9, 76.3, 75.6, 55.4, 53.0, 36.6, 36.2, 33.8, 32.2, 23.6, 23.4, 20.9. 3 is identical to aspterric acid (AA) as reported 24,25 . Protein expression, purification and biochemical assay To express and purify pDHAD, primers pDHAD-pET-F and pDHAD-pET-R were used to amplify a 1.7 kb DNA fragment containing pdhad (AT3G23940). The PCR product was cloned into pET28a using NheI and NotI restriction sites. The resulting plasmid pDHAD-pET was transformed into E.coli BL21 (DE3) to give TY08. To express and purify fDHAD (XP_001208445.1), primers fDHAD-pET-F and fDHAD-pET-R were used to amplify a 1.6 kb DNA fragment containing fdhad. The PCR product was cloned into pET28a using NdeI and NotI restriction sites. The resulted plasmid fDHAD-pET was transformed into E. coli BL21 (DE3) to obtain TY09. To express and purify AstD (XP_001213593.1), primers AstD-pET-F and AstD-pET-R were used to amplify a 1.6 kb DNA fragment containing astD. The PCR product was cloned into pET28a using NdeI and NotI restriction sites. The resulted plasmid AstD-pET was transformed into E. coli BL21 (DE3) to obtain TY10. All DHADs fused a 6×His-tag with a molecular weight ~62 kD were expressed at 16°C 220 rpm for 20 h after 100 μM IPTG induction (IPTG was added when OD600 = 0.8). Cells of 1 L culture were then harvested by centrifugation at 5000 rpm at 4°C. Cell pellet was resuspended in 15 mL Buffer A10 (20 mM Tris-HCl pH 7.5, 50 mM NaCl, 8% glycerol, 10 mM imidazole). The cells were lysed by sonication, and the insoluble material was sedimented by centrifugation at 16000 rpm at 4°C. The protein supernatant was then incubated with 3 mL Ni-NTA for 4 h with slow, constant rotation at 4°C. Subsequently the Ni-NTA resin was washed with 10 column volumes of Buffer A50 (Buffer A + 50 mM imidazole). For elution of the target protein, the Ni-NTA resin was incubated for 10 min with 6 mL Buffer A300 (Buffer A + 300 mM imidazole). The supernatant from the elution step was then analyzed by SDS-PAGE together with the supernatants from the other purification steps. The elution fraction containing the recombinant protein was buffer exchanged into storage buffer (50 mM Tris-HCl pH 7.2, 50 mM NaCl, 10 mM MgCl2, 10% glycerol, 5 mM DTT, 5 mM GSH). In vitro activity assays were carried out in 50 μL reaction mixture containing storage buffer, 10 mM (±)-sodium α,β-dihydroxyisovalerate hydrate (4) and 0.5 μM of purified DHAD enzyme. The reaction was initiated by adding the enzyme. After 0.5 h incubation at 30°C, the reactions were stopped by adding equal volume of ethanol. Approximately 0.1 volume of 100 mM phenylhydrazine (PHH) was added to derivatize the product 3-methyl-2-oxo-butanoic acid (5) into 6 at room temperature for 30 min. 20 μL of the reaction mixture was subject to LC-MS analysis. The area of the HPLC peak with UV absorption at 350 nm were used to quantify the amount of 6. (Extended Data Fig. 2). The inhibition percentage of AA on DHADs determined using in vitro biochemical assays are calculated by following equation: inhibition percentage = 1 - initial reaction rate with AA initial reaction rate without AA Growth inhibition assay of S. cerevisiae on plates or in the tubes S. cerevisiae was grown in isoleucine, leucine and valine (ILV) dropout media (20 g/L glucose, 0.67 g/L Difco™ Yeast Nitrogen Base w/o amino acids, 18 mg/L adenine, arginine 76 mg/L, asparagine 76 mg/L, aspartic acid 76 mg/L, glutamic acid 76 mg/L, histidine 76 mg/L, lysine 76 mg/L, methionine 76 mg/L, phenylalanine 76 mg/L, serine 76 mg/L, threonine 76 mg/L, tryptophan 76 mg/L, tyrosine 76 mg/L) to test growth inhibition of AA on S. cerevisiae. S. cerevisiae was incubated at 28°C until OD600 of the control strain without AA treatment reached about 0.8. The ratio of yeast OD600 in media with AA treatment to yeast OD600 in media without AA was calculated as the percentage of growth inhibition. The inhibition curve was plotted as percentage of inhibition versus AA concentrations. To further prove AA affects BCAA biosynthesis, isoleucine, leucine and valine was also complemented to the media with or without treatment of AA. The growth curves of TY05, TY06 and TY07 were also plotted in Extended Data Fig. 4. The OD600 was recorded for every 20 min over a total of 50 h. Percent inhibition. The growth inhibition percentage of AA on S. cerevisiae strain is calculated by dividing the cell density (OD600) of the AA-treated strain to the corresponding untreated strains when OD600 reaches ~ 0.8 using following equation: growth inhibition percentage = 1 - OD 600 of AA treated strain 0.8 in which 0.8 is the OD600 of untreated strain. Growth inhibition assay of plants on plates or in the tubes MS (2.16 g/L Murashige and Skoog basal medium, 8 g/L sucrose, 8 g/L agar) media was used to test the growth inhibition of AA on A. thaliana, Solanum lycopersicum, and Zea mays. A. thaliana, S. lycopersicum, and Z. mays were grown under long day condition (16/8 h light/dark) using cool-white fluorescence bulbs as the light resource at 23°C. AA was dissolved in ethanol and added to the media before inoculating strains or growing plants. The media of control treatment contains the same amount of ethanol, but without AA. Plant growth inhibition assay by spraying AA was firstly dissolved in ethanol and then added to solvent (0.06 g/L Finale® Bayer Inc. + 20 g/L EtOH). The control plants were treated with solvent containing ethanol only. A. thaliana that are resistant to glufosinate (containing the bar gene) were grown under long day condition (16/8 h light/dark) using cool-white fluorescence bulbs as the light resource at 23°C. Spraying treatments began upon the seed germination, and was repeated once every two days with approximately 0.4 mL AA solution per time per pot. Structure determination of holo-pDHAD The gene encoding pDHAD (residues 35–608) was cloned into pET21a derivative vector pSJ2 with an eight histidine (8×His) tag and a TEV protease cleavage site at the N-terminus. The following primers were used for cloning: the forward primer DHAD-F and the reverse primer DHAD-R. The double mutant K559A/K560A for efficient crystallization was designed using the surface entropy reduction prediction (SERp) server 34 . Mutations were generated by PCR using the forward primer K559AK560A-F and reverse primer K559AK560A-R. All constructed plasmids were verified by DNA sequencing. pDHAD purified under aerobic conditions was found to contain no iron-sulfur cluster (apo form). Hence we performed [2Fe-2S] Cluster reconstitution under the atmosphere of nitrogen in an anaerobic box. The protein was incubated with FeCl3 at the ratio of 1:10 for 1 h on ice and then 10 equivalents of Na2S per protein was added drop-wise every 30 min for 3 h. The reaction mixture was then incubated overnight. Excess FeCl3 and Na2S were removed using a SephadexTM G-25 Fine column (GE Healthcare) 26 . The reconstituted holo-pDHAD was crystallized in an anaerobic box. The proteins (at 10 mg/mL) were mixed in a 1:1 ratio with the reservoir solution in a 50 μL volume of 2 μL and equilibrated against the reservoir solution, using the sitting-drop vapor diffusion method at 16°C. Crystals for diffraction were observed in 0.1 M sodium acetate pH 5.0, 1.5 M ammonium sulfate after 5 d. All crystals were flash-cooled in liquid nitrogen after cryo-protected with solution containing 25% glycerol, 1.5 M ammonium sulfate, 0.1 M sodium acetate pH 5.0. The data were collected at 100K and at the Beam Line 19U1 in Shanghai Synchrotron Radiation Facility (SSRF). Diffraction data of holo-pDHAD was collected at the wavelength of 0.97774 Å. The best crystals diffracted to a resolution of 2.11 Å. The Ramachandran plot favored (%), allowed (%) and outlier (%) are 98.05, 1.60, and 0.36 respectively. All data sets were indexed, integrated, and scaled using the HKL3000 package 35 . The crystals belonged to space group P42212. The statistics of the data collection are summarized in Extended Data Table 1. The holo-pDHAD structure was solved by the molecular replacement method Phaser embedded in the CCP4i suite and the L-arabinonate dehydratase crystal structure (PDB_ID: 5J83) as the search model. All the side chains were removed during the molecular replacement process 36,37 . The resulting model were refined against the diffraction data using the REFMAC5 program of CCP4i 38 . Based on the improved electron density, the side chains of holo-pDHAD protein, iron sulfur cluster, water molecule, acetate ion, sulfate ions, and magnesium ion were manually built using the program WinCoot 39 . The R work and R free values of the structure are 17.67% and 22.15%, respectively. The detailed refinement statistics are summarized in Extended Data Table 1. The geometry of the model was validated by WinCoot. Structural factor and coordinate of holo-pDHAD have been deposited in the Protein Bank (PDB code: 5ZE4). Homology modelling of AstD and docking of substrate or AA into active site of holo-pDHAD The structure of holo-pDHAD was prepared in Schrodinger suite software under OPLS3 force field 40 . Hydrogen atoms were added to reconstituted crystal structures according to the physiological pH (7.0) with the PROPKA tool in Protein Preparation tool in Maestro to optimize the hydrogen bond network 26,41 . Constrained energy minimizations were conducted on the full-atomic models, with heavy atom coverage to 0.5 Å. The homology model was performed in Modeller 9.18 42 , using the crystal structure of holo-pDHAD solved in this work as a template. Sequence alignment in Modeller indicated that AstD and pDHAD shared 56.8% sequence identity and 75.0% sequence similarity (Extended Data Fig. 6). All the highly conserved residues and motifs were properly aligned. A total of 2000 models were generated for each target in Modeller with the fully annealed protocol. The optimal models were chosen for docking studies according to DOPE (Discrete Optimized Protein Energy) score. All ligand structures were built in Schrodinger Maestro software 26 . The LigPrep module in Schrodinger software was introduced for geometric optimization by using OPLS3 force field 40 . The ionization state of ligands were calculated with Epik tool employing Hammett and Taft methods in conjunction with ionization and tautomerization tools 43 . The docking of a ligand to the receptor was performed using Glide 44 . We included cofactors observed in crystal structure during the docking. Since both water and SO4 2− occupied the catalytic site, they were excluded prior to docking. Cubic boxes centered on the ligand mass center with a radius 8 Å for all ligands defined the docking binding regions. Flexible ligand docking was executed for all structures. Ten poses per ligand out of 20,000 were included in the post-docking energy minimization. The best scored pose for the ligand was chosen as the initial structure for further study. The MM/GBSA method was introduced to evaluate the ligand binding affinity based on the best scored docking pose in Schrodinger software. Figures are prepared in PyMOL and Inkscape 45,46 . Both of native substrate α,β-dihydroxyisovalerate and AA were docked into the catalytic site of pDHAD. The cross-section electrostatic surface map shows this unique catalytic pocket has a positively charged internal and a hydrophobic entrance, which binds to negatively charged “head” and hydrophobic “tail” of substrate or AA respectively. Thus the negatively charged “head” can lead both of the substrate and AA into the catalytic chamber. The bulky hydrophobic tricyclic moiety of AA, however, provides stronger hydrophobic interactions to the entrance and blocks the entrance of active site due to the hydrophobic residues at the entrance, including G68, A71, I72, I134, A133, M141, V212, F215, M498 and P501. In contrast, the smaller “tail” of native substrate provides less interactions to entrance because the smaller size limits efficient hydrophobic contact to nearby residues. This implies that once AA binds to pDHAD, it can prevent substrate approaching the active site. We also introduced molecular mechanics generalized Born and surface area (MM/GBSA) continuum solvation method, an widely used approach for relative binding energy calculation, to evaluate the relative binding affinity for both ligands 47 . The MM/GBSA calculations had been done in Prime 48 (Schrödinger 2015 suite). The MM/GBSA energy was calculated using following equation: Δ G bind = E complex - E protein - E ligand E denotes energy and includes terms such as protein–ligand van der Waals contacts, electrostatic interactions, ligand desolvation, and internal strain (ligand and protein) energies, using VSGB2.0 implicit solvent model with the OPLS2005 force field. The solvent entropy is also included in the VSGB2.0 energy model, as it is for other Generalized Born (GB) and Poison–Boltzmann (PB) continuum solvent models. MM/GBSA calculation shows that the relative binding energy for AA and α,β-dihydroxyisovalerate is −18.6 ± 0.3 kcal/mol and −13.3 ± 0.2 kcal/mol respectively, which shows the binding constant of AA to active site is about 6000 times greater than α,β-dihydroxyisovalerate. This further confirms that AA is a competitive inhibitor of pDHAD. Cytotoxicity assay of AA Cell proliferation experiments were performed in a 96-well format (five replicates per sample) using melanoma cell line A375 and SK-MEL-1. AA treatments were initiated 24 h postseeding for 72 h, and cell survival was quantified using CellTiter-GLO assay (Promega). Cross experiment of A. thaliana To make male sterile A. thaliana, AA was added to chemical hybridization agent (CHA) formulation (250 μM AA, 2% ethanol, 0.1% Tween-80, 1% corn oil in water), which has less inhibition effect on the growth of A. thaliana. Flowers of the AA treated col-0 were selected as the female parent. The non-treated A. thaliana containing a glufosinate resistant gene were used as male parent to donate pollen. 2-week old F1 progeny resulting from the cross were treated by Finale (11.3% glufosinate-ammonium) at 1:2000 dilution. The results are summarized in Extended Data Fig. 9. Construction of the transgenic plants The coding sequence of AstD was codon optimized for A. thaliana. A chloroplast localization signal (CLS) of 35-amino acid residues derived from the N-terminal of A. thaliana DHAD (MQATIFSPRATLFPCKPLLPSHNVNSRRPSIISCS) was fused to N-terminus of the codon optimized AstD. A 3×FLAG-tag was inserted between the CLS and the codon optimized AstD (Supplementary Table 6). The gene block containing CLS, FLAG-tag and astD was synthesized and then cloned into pEG202 vector using Gateway LR Clonase II Enzyme Mix (ThermoFisher scientific). The original CaMV 35S promoter of pEG202 was substituted by Ubiquitin-10 promoter to drive the expression of AstD. The construct was electro-transformed into Agrobacterium tumefaciens strain Agl0 followed by A. thaliana transformation using the standard floral dip method 49 . The A. thaliana Col-0 ecotype was transformed. Positive transgenic plants were selected using the glufosinate resistance marker, and were tested for survival in presence of AA. Protein expression verification with western blot Approximately 0.5 gram of leaf tissue of transgenic A. thaliana was grounded in liquid nitrogen. Proteins were homogenized in 2× SDS buffer followed by 5-min centrifuge at 21,000 g to remove undissolved debris. The supernatant containing resolved proteins were loaded onto a 4–12% Bis-Tris gel, and separated using MOPS running buffer. Transfer was conducted using iBlot2 dry transfer device and PVDF membrane. The total proteins were stained with Ponceau to demonstrate equal loading. Western blotting was performed using Sigma monoclonal anti-FLAG M2-Peroxidase antibody, followed by detection using Amersham ECL Prime detection reagent. Data availability The data that support the findings of this study are available within the paper and its Supplementary Information, or are available from the corresponding authors upon reasonable request. Extended Data Extended Data Figure 1 The rationale of resistance-gene directed discovery of a natural herbicide with new mode of action a, Phylogenetic tree of DHAD among bacteria, fungi and plants. The evolutionary history was inferred by using the Neighbor-Joining method (MEGA7). Units represent the number of amino acid substitutions per site. b, Representatives of small molecules that inhibit DHAD in vitro, but fail to inhibit plant growth. c, Examples of co-localization of biosynthetic gene clusters (BGCs) and targets. The biosynthetic core genes are shown in blue and the self-resistance enzymes (SREs) are shown in red. The blockbuster cholesterol-lowering lovastatin drug targets HMG-CoA reductase (HMGR) in eukaryotes. In the fungus Aspergillus terreus that produces lovastatin, a second copy of HMGR encoded by ORF8 is present in the gene cluster (top). BGC of the immunosuppressant mycophenolic acid from Penicillium sp. contains a second copy of inosine monophosphate dehydrogenase (IMPDH), which represents the SRE to this cluster (bottom). Extended Data Figure 2 Biochemical assays of DHAD functions a, Assaying DHAD activities in converting the dihydroxyacid 4 into the α-ketoacid 5. Formation of 5 can be detected on HPLC by chemical derivatization using phenylhydrazine (PHH) to yield 6. b, LC-MS traces of the biochemical assays of A. thaliana DHAD (pDHAD). Extracted ion chromatogram (EIC) of positive ion mass of [M+H]+=207 is shown in red. i. The derivatization reaction was validated by using the authentic 5. ii. The bioactivity of pDHAD in converting 4 into 5 was validated. iii. Addition of DMSO to pDHAD enzymatic reaction mixture has no effect. iv. Addition of 10 μM AA to the reaction mixture abolished pDHAD activity. The experiments were repeated independently for 3 times with similar results. Extended Data Figure 3 Inhibition assay of different DHADs using AA Three DHAD enzymes were assayed, including pDHAD ( p lant DHAD from A. thaliana), fDHAD ( f ungal housekeeping DHAD from A. terreus) and AstD (DHAD homolog within ast cluster). IC50 and K i values of AA were measured based on inhibition percentage at different AA concentrations. Center values are averages, errors bars are s.d.; n = 3 biologically independent experiments. a, Plot of the inhibition percentage of 0.5 μM fDHAD as a function of AA concentration. b, Plot of the inhibition percentage of 0.5 μM pDHAD as a function of AA concentration. c, Plot of the inhibition percentage of 0.5 μM AstD as a function of AA concentration. d, Analysis of inhibitory kinetics of AA on pDHAD using the Lineweaver-Burk method at different concentrations of AA (left). Linear fitting of apparent Michaelis constant (K M,app) as a function of AA concentration yields the inhibition constant (K i) of AA on pDHAD (right). Extended Data Figure 4 Growth curve of S. cerevisiae ΔILV3 expressing AstD and fDHAD The genome copy of DHAD encoded by ILV3 was first deleted from Saccharomyces cerevisiae strain DHY ΔURA3 to give UB02. UB02 was then either chemically complemented by growth on ILV (leucine, isoleucine and valine)-containing media or genetically by expressing of fDHAD or AstD episomally (TY06 or TY07, respectively). The empty vector pXP318 was also transformed into UB02 to generate a control strain TY05. The optical density of cell growth under different conditions were plotted as a function of time. Center values are averages, errors bars are s.d.; n = 3 biologically independent experiments. a, Growth curve in ILV dropout media with no AA. b, Growth curve in ILV dropout media with 125 μM AA. c, Growth curve in ILV supplemented media; d, Growth curve in ILV supplemented media with 250 μM AA. Extended Data Figure 5 X-ray Structure of holo-pDHAD and homology model of AstD a, Superimpositions monomer of holo-pDHAD (PDB: 5ZE4, 2.11 Å) and RlArDHT (PDB: 5J84). The holo structure containing the 2Fe-2S cofactor and Mg2+ ion in the active site. The structure of holo-pDHAD is in white; the crystal structure of RlArDHT is in cyan. b, Superimpositions of holo-pDHAD and homology modeled AstD. The structure of AstD was constructed by homology modeling based on the structure of holo-pDHAD. The structure of holo-pDHAD is in white; the crystal structure of AstD is in green. c, The electron density map of cofactors in the holo structure of pDHAD. White grid: 2Fo-Fc map at 1.2 σ level. Green grid: Fo-Fc positive map at 3.2σ level. Cyan sticks: acetic acid molecule. d, Comparison of the active sites in the crystal structure of pDHAD and the modeled structure of AstD. The cartoon represents superimposed binding sites of pDHAD (white) and AstD (green). The shift of a loop in AstD, where L518 (correspond to V496 in pDHAD) is located, coupled with a larger L198 residue (correspond to I177 in pDHAD) lead to a smaller hydrophobic pocket of AstD than that in pDHAD. e, The surface of binding sites of AstD (left) and pDHAD (right). The smaller hydrophobic channel in modeled AstD cannot accommodate the AA molecule (yellow balls-and-sticks). Extended Data Figure 6 Sequence alignment between pDHAD and AstD The sequence identity between pDHAD and AstD is 56.8%, whereas the similarity between them is 75.0%. Residues were colored according to their property and similarity. Extended Data Figure 7 Spray assay of AA on A. thaliana Glufosinate resistant A. thaliana was treated with (right) or without (left) AA in the solvent, which is a commercial glufosinate based herbicide marketed as Finale®. To improve the wetting and penetration, AA was firstly dissolved in ethanol and then added to solvent (0.06 g/L Finale® Bayer Inc. + 20 g/L ethanol) to make 250 μM AA spraying solution. The control plants were treated with solvent containing ethanol only. Spraying treatments began upon the seeds germination, and were repeated once every two days with approximately 0.4 mL AA solution per time per pot for 4 weeks. The picture shown below is taken after one month of treatment. The application rate of AA is approximately 1.6 lb/acre, which is comparable to the commonly used herbicide glyphosate (0.75~1.5 lb/acre). The experiments were repeated independently for 3 times with similar results. Extended Data Figure 8 Specific inhibition of anther development in A. thaliana Comparison of flower organs between the AA treated (a–c) and non-treated (d–f) Arabidopsis. a compare to d, the AA treated flower shows abnormal pistil elongation due to the lack of pollination. b compare to e, the AA treated flower is missing one stamen. c compare to f, the AA treated anther is depleted of healthy and mature pollen. The experiments were performed twice with similar results. Extended Data Figure 9 Schematic illustration of results from the cross experiment a, Wild type A. thaliana treated with 250 μM AA was pollinated with pollen from the un-treated plant that carries the glufosinate resistant gene. Offspring was obtained, and inherited the glufosinate resistance from the pollen donor. b, similar as in a, except that the pollen donor was also treated with 250 μM AA. No offspring was obtained from this cross. Similar results were obtained with the treatment of AA at 100 μM. Extended Data Table 1 Data collection and refinement statistics (molecular replacement). holo-pDHAD Data collection Space group P42212 Cell dimensions   a, b, c (Å) 135.5, 135.5, 66.0  α, β, γ (°) 90, 90, 90 Resolution (Å) 47.89-2.11 (2.15-2.11)* R sym or R merge 0.189 (1.240) I/σI 17.86 (2.33) Completeness (%) 100 (100) Redundancy 25.1 (23.1) Refinement Resolution (Å) 95.79-2.11 No. reflections 33235 (1714) R work/R free 0.1767/0.2216 No. atoms  Protein 4208  Ligand/ion 24  Water 118 B-factors  Protein 26.60  Ligand/ion 46.53  Water 26.22 R.m.s. deviations  Bond lengths (Å) 0.007  Bond angles (°) 1.195 * Values in parentheses are for highest-resolution shell. Supplementary Material 1 2