Total Biosynthesis for Milligram-Scale Production of Etoposide Intermediates in a Plant Chassis

Abstract

Etoposide is a plant-derived drug used clinically to treat several forms of cancer. Recent shortages of etoposide demonstrate the need for a more dependable production method to replace the semisynthetic method currently in place, which relies on extraction of a precursor natural product from Himalayan mayapple. Here we report milligram-scale production of (−)-deoxypodophyllotoxin, a late-stage biosynthetic precursor to the etoposide aglycone, using an engineered biosynthetic pathway in tobacco. Our strategy relies on engineering the supply of coniferyl alcohol, an endogenous tobacco metabolite and monolignol precursor to the etoposide aglycone. We show that transient expression of 16 genes, encoding both coniferyl alcohol and main etoposide aglycone pathway enzymes from mayapple, in tobacco leaves results in the accumulation of up to 4.3 mg/g dry plant weight (−)-deoxypodophyllotoxin, and enables isolation of high-purity (−)-deoxypodophyllotoxin after chromatography at levels up to 0.71 mg/g dry plant weight. Our work reveals that long (>10 step) pathways can be efficiently transferred from difficult-to-cultivate medicinal plants to a tobacco plant production chassis, and demonstrates mg-scale total biosynthesis for access to valuable precursors of the chemotherapeutic etoposide.

Graphical Abstract

The topoisomerase II inhibitor etoposide is used in chemotherapy regimens for the treatment of lung cancer, testicular cancer, lymphomas, and other malignancies.¹ Currently, etoposide is produced semisynthetically from (−)-podophyllotoxin extracted from the plant Sinopodophyllum hexandrum.^2,3 However, despite being on the World Health Organization’s list of essential medicines,⁴ there have been several etoposide shortages in the past decade⁵ and S. hexandrum is considered endangered.⁶ One alternative approach to the current production method is to isolate (−)-podophyllotoxin from plant tissue culture systems; such methods have been attempted with some success,⁷ but these techniques are technically challenging and not easily scalable.⁸ Recently, we reported the elucidation of the complete biosynthetic pathway to the etoposide aglycone (EA),⁹ a more direct precursor to etoposide than (−)-podophyllotoxin (Scheme 1, Figure S1), thereby opening up potential for a metabolic engineering approach to etoposide production. This approach could decrease the dependence on environmental factors¹⁰ and would allow for more facile access to analogues of the aglycone scaffold in a manner complementary to current synthetic and chemoenzymatic methods.¹¹ Additionally, this approach is potentially more efficient and less labor intensive than other methods given the functional equivalence of total biosynthesis to a one-pot chemical synthesis.

Scheme 1. Engineered Biosynthetic Pathway for Production of EA and Late-Stage Pathway Intermediates^a.

^aArrows represent previously characterized biosynthetic genes. The coniferyl alcohol (CA) pathway is shared by all vascular plants, including tobacco, while the (−)-deoxypodophyllotoxin (DPT) and etoposide aglycone (EA) genes are less widespread (see Supporting Information). The latter five DPT enzymes and EA cytochromes P450 are thought to be more specific to EA production and are not found in tobacco.

Only a few complete biosynthetic pathways for plant natural products have been engineered in heterologous hosts, typically yeast, with important examples including artemisinic acid,¹² the benzylisoquinoline¹³ and monoterpene indole alkaloids,¹⁴ and cannabinoids.¹⁵ Many of these efforts show promise for alleviating supply problems associated with the native plants. While yeast is appealing because of its scalability, engineering plant biosynthetic pathways in this organism remains challenging.¹⁶ It is estimated that engineering the four-step artemisinic acid pathway into yeast required roughly 150 person-years to complete, from discovery through optimization.¹⁷ Recently, the wild relative of tobacco Nicotiana benthamiana (referred to here as tobacco) has emerged as a platform for plant biosynthetic pathway discovery and engineering due to its ability to transiently express enzymes quickly via infiltration of Agrobacterium tumefaciens (Agro.) strains harboring enzyme-encoding genes as part of a rapid design−build−test cycle.^9,18 Plant platforms may be advantageous over microbial platforms because plant enzymes, particularly cytochromes P450 (CYP), are often difficult to express in microbial hosts.¹⁶ Additionally, different plant species typically share the same subcellular compartments, cofactors, and metabolic precursors, many of which are absent in microbes, simplifying pathway transfer.¹⁸ In one recent example,¹⁹ Reed et al. demonstrated β-amyrin production in tobacco at the level of 3.3 mg/g dry leaf weight (DW) with a two-enzyme system, sufficient for isolation of near gram-scale quantities of the product from ∼460 plants. When coupled with expression of individual cytochromes P450, a variety of derivatives were produced at similar yields and tested for antiproliferative and anti-inflammatory activity in human cell lines, an exciting proof of concept that tobacco can be used for production of drug candidates difficult to access using traditional synthetic methods at the scale required for evaluation of their medicinal potential.

Precursor supply engineering has classically enabled high-level production of metabolites in microorganisms but has been widely unexplored in plants, the most notable exception being in terpene biosynthesis.^19,20 Previously we reported production of (−)-deoxypodophyllotoxin (DPT), a late-stage intermediate in the EA pathway, in tobacco at a yield of 11.4 ± 3.8 μg/g DW from co-expression of the eight DPT pathway genes by Agro.-infiltration.⁹ Although products at these levels can be readily distinguished by mass spectrometry, a large increase would be required for facile isolation of pure DPT and other late-stage intermediates. Here, we report that by co-expressing eight precursor supply genes found in all vascular plants with eight other biosynthetic genes required for DPT production (Scheme 1), yields up to 4.3 mg/g DW can be achieved.

We began our efforts to increase the yield of the EA pathway in tobacco by determining the key pathway bottleneck. Previously, our work had shown that direct exogenous addition of (+)-pinoresinol to the plant chassis resulted in an 8-fold improvement in DPT yield.⁹ Therefore, we hypothesized that the laccase-catalyzed, dirigent protein (DIR)-directed coniferyl alcohol (CA) dimerization that generates (+)-pinoresinol was the yield-limiting step. This dimerization is thought to occur in the plant cell apoplast,²¹ necessitating the transport of metabolites out of and back into the cytoplasm. To test if CA coupling limits pathway yield, we infiltrated CA into tobacco leaves 4 days after infiltration of a mixture of Agro. strains harboring the DPT pathway genes, harvesting leaves 1 day later. This resulted in a 13-fold increase in DPT production, as quantified by liquid chromatography-mass spectrometry (LC-MS), compared to a parallel experiment where no CA was infiltrated (Figure 1). Notably, when the Agro. strain harboring the DIR gene was replaced with a strain harboring a GFP-encoding gene, the yield increase dropped to 2-fold. This suggests that while tobacco is able to produce (+)-pinoresinol endogenously, over-expression of the stereochemistry-mediating dirigent protein is important for limiting flux into off-target pathways.

Figure 1.

Impact of CA infiltration and DIR expression on DPT production in planta according to LC-MS of extracts from Agro.- infiltrated tobacco. Gray bar height indicates mean (error bars show ± SD); individual biological replicates are represented as dots. Colors indicate absence or presence of CA infiltration (−CA or +CA, respectively).

Because yield increases were only observed when CA was infiltrated, we hypothesized that, in addition to CA coupling, CA supply in the tobacco leaf might also limit pathway flux. A biological means for increasing the CA pool in leaves is attractive because it avoids the need for infiltration of synthetic CA, a costly and likely inefficient step in a large-scale production process and, furthermore, a strategy that is not broadly applicable to other plant natural products for which precursors may not be able to access the cytoplasm upon infiltration. Thus, candidate CA biosynthetic genes were selected from our previous S. hexandrum RNA-Seq data set⁹ based on homology to characterized Arabidopsis thaliana genes and co-expression with DPT pathway genes (Tables S1 and S2). The CA biosynthetic pathway has been extensively studied and characterized, and the most widely accepted pathway includes nine steps catalyzed by eight different enzymes: PAL, C4H, 4CL, HCT, C3H, CCoA-OMT, CCR, and CAD (Scheme S1).²² The top co-expression candidate for each was selected, and these genes were incorporated into plasmids in separate strains of Agro. When these strains were co-infiltrated alongside the others harboring the DPT pathway, a 680-fold increase in DPT yield was observed compared to the GFP control (Figure S2). The yield for this initial attempt was approximately 3.5 ± 1.2 mg/g DW, compared to 5.2 ± 0.6 μg/g DW in the non-engineered control.

To investigate the impacts of individual CA biosynthetic genes on the increased yield, we examined yields after sequential addition of each gene to the pathway. From this analysis, it was determined that PAL plays an important role, with PAL expression being responsible for a 5- to 6-fold increase in yield (Figures 2 and S3). Additionally, CCoA-OMT, CCR, and CAD were found to collectively be responsible for a 95- to 160-fold yield increase, though the individual contributions of each enzyme were not consistent between different experimental batches (Figures 2 and S3–S5). While the other four enzymes do not appear to increase flux significantly based on sequential addition, replacing these four with GFP diminished yields significantly (Figure S6). In certain experiments, it appears that a subset of these eight genes is sufficient to generate the overall yield increase; however, the most consistent increase is observed when all eight are included (Figures S7 and S8). Interestingly, PAL does not appear to have much impact on yield when expressed alone with the DPT pathway (Figure S8), suggesting the intermediary steps that connect the resulting cinnamic acid to CA are necessary for the increased production.

Figure 2.

Impact of individual CA pathway enzyme expression on DPT yield in planta according to LC-MS of extracts from Agro.- infiltrated tobacco. Gray bar height indicates mean (error bars show ± SD); individual biological replicates are represented as dots. Asterisks indicate a significant difference between sets, p < 0.05. Gray-shaded boxes indicate enzymes expressed via Agro.-infiltration, with numbers showing the amount of the GFP-expressing strain used relative to the other strains.

To further improve DPT yields in the tobacco system, total Agro. mixture OD₆₀₀ and infiltration-to-harvest time were both optimized. It was determined that a total Agro. mixture OD₆₀₀ of 3.0 and an infiltration-to-harvest time of 7−9 days were optimal, yielding up to 4.3 mg/g DW (Figures S9 and S10). We next used the optimized conditions and scaled production up to 15−20 plants. High-purity DPT was successfully isolated at yields up to 0.71 mg/g DW using silica gel flash chromatography and preparative HPLC (Table S3A).

Having achieved yield boosts enabling isolation and NMR characterization of DPT, we next investigated the applicability of CA pathway over-expression in tobacco for EA and EA analogue biosynthesis. In our previous work, EA was only detected in planta when the full biosynthetic pathway was expressed along with exogenous addition of the precursor (−)-matairesinol.⁹ Here we observe that, with CA pathway expression, EA can now be detected without precursor infiltration (Figure 3A).

Figure 3.

EA and (−)-morelensin production. (A) LC-MS extracted ion chromatograms (EIC) at 401 m/z, [EA + H]⁺. Green asterisk indicates EA; other peaks are likely derived from in-source fragmentation of EA glycosides. Both traces are from experiments where noted genes were expressed in tobacco via Agro.-infiltration. (B) Engineered pathway for (−)-morelensin production. Dotted arrows depict omitted enzymes. (C) (−)-Morelensin production with exogenous addition of varying CA levels or co-infiltration with CA pathway (PW)-expressing Agro. strains. Gray bar height indicates mean (error bars show ± SD); individual biological replicates are represented as dots. Asterisks indicate a significant difference between sets, p < 0.05.

Given the high levels of EA produced, we next tested whether EA analogues not previously reported to accumulate in mayapple could be made using our engineered pathway. One potential bottleneck is the stereospecific C−C bond formation step catalyzed by 2-ODD. We had previously tested 2-ODD substrate specificity with metabolites other than (−)-yatein.⁹ These tests showed that the enzyme can also utilize (−)-bursehernin, indicating flexibility in processing alternative substrates. Here, by co-expressing the entire CA pathway and a subset of the DPT pathway enzymes lacking CYP71CU1 and OMT1 (Figure 3B), 2.2 mg/g DW yield (1.4 mg/g DW isolated yield) of the alternative 2-ODD product, (−)-morelensin, was obtained in planta without exogenous precursor addition (Figure 3C, Table S3B). Importantly, this yield was significantly higher than what could be obtained by exogenous addition of a saturated CA solution (2.5 mM), supporting the value of engineering the production of CA in planta. We further report here that the EA P450 enzymes, CYP82D61 and CYP71BE54, catalyze hydroxylation and demethylation, respectively, of (−)-morelensin, producing (−)-4′-demethyl-5′-desmethoxy-epipodophyllotoxin (5′-desmethoxy-EA, not known to accumulate in mayapple) in planta. Co-expression of the CA pathway resulted in an order of magnitude yield increase of 5′-desmethoxy-EA compared to exogenous CA addition (Figures S11 and S12).

In this work, we have demonstrated the applicability of precursor supply engineering for high-yield plant natural product biosynthesis in tobacco. By co-expressing eight genes that enable CA production from intracellular phenylalanine alongside eight genes that generate DPT from CA—to our knowledge, the longest natural product pathway reconstituted in a plant heterologous host to date—a 2 orders of magnitude DPT yield increase was observed compared to when the CA pathway genes were excluded. Additionally, we demonstrated a significant increase in the yield of EA from the system and showed how this method allows for access to analogues of etoposide pathway intermediates.

We anticipate this work will be important for development of tobacco as a platform for medicinal plant natural product production. By our count, 10% of WHO essential medicines are plant natural products or derivatives,⁴ and a majority of these medicines still rely on the native plant for production (e.g., etoposide, Taxol²³). Our work shows that by rewiring central metabolism in situ to meet the demands of a given biosynthetic pathway, tobacco transient expression can be used to obtain isolatable milligram-scale quantities of natural plant small molecules and biosynthetically accessible analogues.