Full-length genes of maize ZmCPS3 (Zm00001d024512) and ZmCPS4 (Zm00001d048874) were synthesized with support of a Department of Energy Joint Genome Institute Community Science Program grant (CSP#2568). Genes were inserted into the second multiple cloning site of a pACYC-Duet plasmid also carrying the GGPP synthase from Abies grandis (Morrone et al., 2010) to form the constructs pACYC-Duet:AgGGPPS-ZmCPS3 and pACYC-Duet:AgGGPPS-ZmCPS4 (Supplementary Table S1). Additional constructs used in co-expression assays were described previously, including pACYC-Duet:AgGGPPS-ZmAN2 (Zm00001d029648) (Z. mays ent-CPP synthase), pET28b:ZmKSL3 (Zm00001d002349) (Z. mays ent-kaurene synthase), pET28b:ZmKSL4 (Zm00001d032858) (Z. mays dolabradiene synthase), pET28b:GrTPS1 (Grindelia robusta LPP synthase), and pET15b:MvELS (Marrubium vulgare 9,13-epoxy labd-14-ene synthase; Zerbe et al., 2013, 2014; Mafu et al., 2018).

Functional co-expression of enzymes was carried out using an engineered Escherichia coli platform for enhanced diterpenoid production (Morrone et al., 2010) as previously described (Mafu et al., 2018). The pACYC-Duet:AgGGPPS-ZmCPS3, pACYC-Duet:AgGGPPS-ZmCPS4, or pACYC-Duet:AgGGPPS-ZmAN2 constructs were expressed alone or in combination with ZmKSL3, or ZmKSL4. GrTPS1 was expressed in combination with pACYC-Duet:AgGGPPS to form LPP and an additional combination with pET15b:MvELS to form manoyl oxide for use as authentic standards. In brief, cultures were grown in 50 mL Terrific Broth (TB) medium to an OD₆₀₀ of ∼0.6 at 37°C. After cooling to 16°C, cultures were induced with 1 mM isopropyl-thio-galactoside (IPTG) and 25 mM sodium pyruvate and incubated for a further 72 h. Enzyme products were extracted with 50 mL of hexane, concentrated under air, and resuspended in 1 mL hexane for GC-MS analysis.

Gas chromatography mass spectrometry analysis of enzyme products was performed on an Agilent 7890B GC interfaced with a 5977 Extractor XL MS Detector at 70 eV and 1.2 mL min^-1 He flow, using an Agilent HP5-MS column (30 m, 250 μm i.d., 0.25 μm film) with a sample volume of 1 μL and the following GC parameters: pulsed splitless injection at 250 and 50°C oven temperature; hold at 50°C for 3 min, 20°C min^-1 to 300°C, hold 3 min. MS data from 90 to 600 mass-to-charge ratio (m/z) were collected after a 8 min solvent delay. Products were identified using comparison to authentic standards or, where these were not available, comparison to published mass spectra and the National Institute of Standards and Technology (NIST version 2.0) mass spectral library (Agilent).

Diterpenoids were produced via large-scale (12 L) enzyme co-expression cultures as described above. Hexane extracts were dried using rotary evaporation, resuspended in hexane, and purified by silica column chromatography (230–400 mesh, grade 60) using a hexane:ethyl acetate gradient as the mobile phase. Fractions were further purified on an Agilent 1100 series HPLC with diode array UV detector and an Agilent ZORBAX Eclipse Plus-C8 column (4.6 mm × 150 mm, 5 microns) at a 0.5 mL min^-1 flow rate and H₂O/acetonitrile gradient as mobile phase. Product purity was verified using GC-MS analysis as outlined above. Purified products were dissolved in 0.6 mL deuterated chloroform (CDCl₃; Sigma-Aldrich) containing tetramethylsilane (TMS). NMR spectra were acquired at room temperature on a Bruker Avance III 800 spectrometer equipped with a 5 mm CPTCI. Chemical shifts were calculated by reference to known CDCl₃ (¹³C 77.23 ppm, ¹H 7.24 ppm) signals offset from TMS. All spectra were acquired using standard experiments on a Bruker TopSpin 3.2 software, including 1D ¹H and 1D ¹³C spectra (201 MHz).

For analysis of class II diTPS gene and protein abundance, publicly available transcriptome and proteome inventories were investigated that represent a range of organs and tissues at different developmental stages of healthy maize plants (Walley et al., 2016). All samples derive from B73, with the exception of 2 cm tassels, 1–2 mm anthers, and mature pollen (W23 inbred), and 5-days-old primary root (Mo17 inbred) (Walley et al., 2016). Transcript abundance (as fragments per kilobase of transcript per million mapped reads, FPKM) and protein expression levels were retrieved directly from this public resource (Walley et al., 2016), and were scaled by color either individually by gene or absolute across all four genes of interest.

Plant samples used for gene expression analysis were prepared previously (Mafu et al., 2018). Briefly, maize (var. Golden Queen) seed was germinated in the dark on wetted paper for 4 days at 23°C. Seedlings were transferred and grown hydroponically (Schmelz et al., 2001) for 12 days under 16/8 h light/darkness at 28°C, light intensity of 180 μmol photons m^-2 s^-1, and ∼60% relative humidity. A total of 1 mM CuSO₄ or the corresponding water control was added to the hydroponic medium. Root samples were collected at the time points indicated, with three biological replicates per time point, and immediately frozen in liquid nitrogen for further analysis.

Plants samples used for gene expression analysis were derived from a previous study (Mafu et al., 2018). Maize (var. Mo17) plants were grown in the greenhouse for 53 days in individual, 10 L pots and supplemented with 14-14-14 Scotts Miracle Grow fertilizer. Large nodal roots (≥2 mm dia) were punctured with a 0.6 mm dia steel pin at 1 cm intervals and inoculated with 10 μL of 1 × 10⁷ conidia mL^-1 of Fusarium verticillioides (F.v.), Fusarium graminearum (F.g.), or water control at each wound site. To avoid damage to other tissues not undergoing treatment, sampling was limited to roots on the outer edge of the soil, in contact with the vertical plastic pot wall. Root samples were collected after 7 days and immediately frozen in liquid nitrogen before further processing.

Gene expression analysis of ZmCPS3 and ZmCPS4 was performed on the CuSO₄-treated or pathogen-treated root samples described above. Total RNA was isolated as described elsewhere (Kolosova et al., 2004) and cDNA was synthesized using SuperScript III First-Strand Synthesis Kit (Invitrogen) according to manufacturer’s instructions. Transcript abundance was measured using a BioRad C1000 Touch Thermo Cycler interfaced with a CFX96 Real-Time System, and iTaq Universal SYBR Green Supermix (BioRad) according to manufacturer’s protocols. Mean cycle threshold (Ct) values of at least two technical and three biological replicates were normalized using elongation factor EF1α as a validated reference gene (Schmelz et al., 2011), and fold change values were calculated using the 2^-ΔΔCt method. Gene-specific oligonucleotides used for qPCR analysis were ZmCPS3 [forward: 5′-TGACGTGTGGATTGGGAAGG-3′, reverse: 5′-CGCTCTGTCTGGCTCAAAGA-3′], ZmCPS4 [forward: 5′-CTCAGGCCAGCTTAACGAC-3′, reverse: 5′-CCTTGCCGATCCATACGTC-3′], and EF1α [forward: 5′-TGGGCCTACTGGTCTTACTACTGA-3′, reverse: 5′-ACA TACCCACGCTTCAGATCCT-3′].

Protein sequence alignments were generated using the CLCBio software package (Qiagen), followed by manual curation. Maximum likelihood phylogenetic analysis was performed using PhyML-aBayes version 3.0.1 beta (Guindon et al., 2010) with four rate substitution categories, LG substitution model, BIONJ starting tree, and 500 bootstrap repetitions.

The maize genome (B73 RefGen_v4) contains four class II diTPSs (Schmelz et al., 2014), which comprise the known ent-CPP synthases ZmAN1 and ZmAN2 located on chromosome 1 (Bensen et al., 1995; Harris et al., 2005), and the previously uncharacterized ZmCPS3 and ZmCPS4 positioned on chromosomes 4 and 10, respectively. ZmCPS3 and ZmCPS4 share a protein sequence identity of 55% with each other and 46–56% with ZmAN1 and ZmAN2. Phylogenetic analysis placed ZmCPS3 and ZmCPS4 separate from most ent-CPP synthases on a branch primarily consisting of class II diTPSs that produce prenyl diphosphates of specialized metabolism (Supplementary Figure S1). Although this suggested a role of both enzymes in specialized metabolism, the functional diversity of diTPSs in this group did not allow inference of possible ZmCPS3 and ZmCPS4 functions. Therefore, to inform biochemical analyses, we interrogated key active site residues with known impact on class II diTPS product specificity (Xu et al., 2007b; Mann et al., 2010; Jia et al., 2016, 2017; Mafu et al., 2016; Potter et al., 2016a; Pelot et al., 2018; Schulte et al., 2018). Previous studies identified a His-Asn catalytic dyad that is widely conserved among ent-CPP synthases, including ZmAN1 and ZmAN2, and was shown to direct class II diTPS catalysis toward ent-CPP formation (Potter et al., 2014, 2016b; Cui et al., 2015; Pelot et al., 2016). Neither ZmCPS3 nor ZmCPS4 possess a His-Asn catalytic dyad, but instead feature a Leu-Phe residue pair (Figure 2). These residues are consistent with a recently characterized 8,13-CPP synthase from switchgrass (P. virgatum; Pelot et al., 2018). In addition, presence of a Tyr residue in position 497 and 458 of ZmCPS3 and ZmCPS4, respectively, is consistent with known monocot ent-CPP and (+)-CPP synthases, but contrasts known syn-CPP synthases from rice and switchgrass that feature a His residue in this position (Potter et al., 2016a; Pelot et al., 2018). Although insufficient to allow an unambiguous functional annotation, these active site characteristics disfavored an ent-CPP or syn-CPP synthase activity, and supported a possible function of ZmCPS3 and ZmCPS4 as 8,13-CPP or (+)-CPP synthases or related specialized class II diTPSs.

To test the predicted enzyme activity of ZmCPS3, a synthetic full-length gene was co-expressed with a GGPP synthase from A. grandis using an in vivo E. coli expression platform engineered for diterpenoid production (Morrone et al., 2010). In vivo expression of class II diTPS enzymes using this system readily yields dephosphorylated products, presumably due to the activity of E. coli endogenous phosphatases and thus enables direct hexane extraction and analysis of the corresponding diterpene alcohols (Zerbe et al., 2013, 2015; Pelot et al., 2016; Mafu et al., 2018). For clarity, structures depicted below represent the native prenyl diphosphate products, but were detected via GC-MS and NMR analysis as the corresponding alcohols. Expression of ZmCPS3 resulted in a major product (compound 1; Figure 3) with a retention time of 11.25 min and a fragmentation pattern showing dominant mass ions of m/z 137, 257, and 275 that closely matched the mass spectrum of copalol (i.e., dephosphorylated CPP) produced by ZmAN2 (compound 8; Figure 3). Two additional byproducts detected in the ZmCPS3 product profile represented unconverted GGPP substrate (compound 2) and an unidentified non-diterpenoid contaminant (compound 3; Supplementary Figure S2).

Next we defined the stereochemistry of the ZmCPS3 product by co-expressing ZmCPS3 with characterized maize class I diTPSs that display catalytic specificity toward CPP substrates of different stereochemistries. ZmKSL3 converts ent-CPP into ent-kaurene (Fu et al., 2016), whereas ZmKSL4 forms dolabradiene from ent-CPP and pimarane-type diterpene olefins with (+)-CPP as a substrate (Mafu et al., 2018). As controls, we co-expressed ZmAN2 with either ZmKSL3 or ZmKSL4 to generate ent-kaurene (compound 10) and dolabradiene (compound 11), respectively (Figure 3 and Supplementary Figure S2). Co-expression of ZmCPS3 and ZmKSL3 did not result in any detectable class I diTPS product (Figure 3). The combined activity of ZmCPS3 with ZmKSL4 did not yield dolabradiene, but resulted in several pimarane-related products (compounds 4–7), the most abundant of which was identified as pimara-8,14-diene (compound 5) by comparison to reference mass spectra (Supplementary Figures S2, S3). These products are consistent with the previously reported activity of ZmKSL4 with the established (+)-CPP synthase, A. grandis abietadiene synthase variant D621A (Mafu et al., 2018). On the basis of these results, ZmCPS3 was designated as a (+)-CPP synthase.

Escherichia coli co-expression of the full-length, synthetic gene encoding ZmCPS4 with the A. grandis GGPP synthase yielded a major product (compound 12) with a retention time of 11.26 min, indicating a related but distinct compound as compared to (+)-CPP and ent-CPP formed by ZmCPS3 and ZmAN2 in vitro, respectively (Figure 4). Presence of signature mass ions of m/z 275 and 257 in the fragmentation pattern of this product indicated the expected labdane structure, but additional major mass ions of m/z 205 and 149 suggested a structure distinct from the common ent-CPP, (+)-CPP, or syn-CPP products observed in monocot crops (Xu et al., 2004; Harris et al., 2005; Wu et al., 2012). Indeed, the fragmentation pattern of compound 12 matched the product of a recently identified class II diTPS from switchgrass that forms 8,13-CPP (Supplementary Figure S2; Pelot et al., 2018). To further verify the identity of this ZmCPS4 product, nuclear magnetic resonance (NMR) spectroscopy was performed. For this purpose, large-scale E. coli co-expression cultures were used to produce an excess of 1 mg of the product, which was then purified using silica column chromatography and semi-preparative high-pressure liquid chromatography (HPLC). For the purified product, 1D ¹H and ¹³C NMR spectra were acquired and validated the ZmCPS4 product as 8,13-CPP in comparison to published spectra (Supplementary Figure S4; Pelot et al., 2018).

In addition to the primary 8,13-CPP product, a minor, more polar, ZmCPS4 product with a retention time of 11.79 min was also observed (compound 13; Figure 4). This product was identified as labda-13-en-8-ol diphosphate (LPP) based on comparison to the retention time and mass spectrum of LPP produced by a known LPP synthase from G. robusta (GrTPS1; Figure 4; Zerbe et al., 2013). Low abundance of this product prevented further stereochemical analysis via NMR. Additional byproducts (compounds 14–16) present in the ZmCPS4 product profile represent manoyl oxide (compound 15) as based on comparison to a manoyl oxide standard produced from the coupled reaction of G. robusta GrTPS1 and MvELS from M. vulgare (Zerbe et al., 2013; Pateraki et al., 2014), and a closely related unidentified diterpenoid (compound 14; Figure 4 and Supplementary Figure S2). These products are likely resulting from rearrangement of the ZmCPS4 products 8,13-CPP and LPP after dephosphorylation by endogenous E. coli phosphatases, as has been described for various related class II diTPSs (Zerbe et al., 2015; Mafu et al., 2018; Pelot et al., 2018). To test for possible downstream products of 8,13-CPP and LPP, ZmCPS4 was co-expressed with the currently known maize class I diTPS functions, including the ent-kaurene synthase ZmKSL3 and the dolabradiene synthase ZmKSL4 (Fu et al., 2016; Mafu et al., 2018). No new products were detected when co-expressing ZmCPS4 and ZmKSL3 as compared to the expression of ZmCPS4 alone (Figure 4). When combining ZmCPS4 with ZmKSL4, compound 15 identified as manoyl oxide was significantly increased and, albeit at low abundance, an additional product (compound 17) was formed that predictably represents a labdane diterpene olefin as based on characteristic mass fragments of m/z 272 and 257 (Supplementary Figure S2).

To further investigate the functions of ZmCPS3 and ZmCPS4, we next examined their gene expression patterns using publicly available transcriptome and proteome data across various organs, tissues, and developmental stages of unstressed maize plants (Walley et al., 2016). With the exception of 5-days-old primary root (Mo17 inbred) and 2 cm tassels, 1–2 mm anthers, and mature pollen (W23 inbred), all samples were from the B73 genotype (Walley et al., 2016). Detectable transcripts of ZmCPS3 were distributed across all organs and tissues with highest abundance included germinating kernels (Figure 5A). Notably, when specifically analyzing gene expression in roots, transcript of ZmCPS3 was observed to be significantly more abundant with expression levels 5–481 times higher as compared to ZmAN1, ZmAN2, and ZmCPS4 in the same tissue (Figure 5B). Transcript of ZmCPS4 was detected at only trace levels and present exclusively in primary and secondary root and root cortex tissues (Figure 5). Consistent with observed transcript abundance, query of public proteome data (Walley et al., 2016) showed that protein levels of ZmCPS3 were highest in root tissues (Supplementary Figure S5). ZmCPS4 protein was not detected in the analyzed proteome.

Previous studies demonstrated that ZmAN2 gene expression is induced under both pathogen (F.g. and F.v.) and oxidative stress in above- (Harris et al., 2005; Christensen et al., 2018) and below-ground tissues (Mafu et al., 2018). In the context of these findings, the relatively higher transcript abundance of ZmCPS3 in roots and the low but possibly root-specific expression of ZmCPS4 suggested a role of these enzymes in belowground stress responses. To investigate this hypothesis, we analyzed transcript abundance of ZmCPS3 and ZmCPS4 using a previously reported set of samples of 53-days-old maize Mo17 roots incision-inoculated with fungal spores of F.v. and F.g. and harvested 7 days later (Mafu et al., 2018). Plants treated by incision-wounding only were used as controls. QPCR analysis showed no significant fold change in ZmCPS3 transcript abundance in roots exposed to F.v. or F.g. as compared to wounded controls (Figure 6). On average, gene expression of ZmCPS4 was significantly decreased in response to F.v. and F.g. elicitation as compared to control plants.

To examine ZmCPS3 and ZmCPS4 gene expression in response to abiotic stress, qPCR analyses were performed on two-week-old roots of maize (var. Golden Queen) plants that were exposed to 1 mM CuSO₄ treatment (as a proxy for oxidative stress), previously shown to induce diterpenoid biosynthesis (Mafu et al., 2018). All treatments were compared to a water-treated control at each time point using the 2^-ΔΔCt method in which the control has a fold change of 1. Transcript levels of ZmCPS3 in CuSO₄-treated roots did not differ significantly from those observed in the roots of water-treated control plants at 2 and 4 h post treatment, but were significantly reduced at the 24 h post treatment time point (Figure 6) Conversely, ZmCPS4 showed increased transcript abundance in CuSO₄-treated roots as early as 2 h post treatment and with an up to sixfold change at 4 h, before decreasing again after 12 h, with a peak of sixfold increase in transcript abundance (Figure 6).