A microbial supply chain for production of the anti-cancer drug vinblastine

Monoterpene indole alkaloids (MIAs) are a diverse family of complex plant secondary metabolites with many medicinal properties, including the essential anti-cancer therapeutics vinblastine and vincristine1. As MIAs are difficult to chemically synthesize, the world’s supply chain for vinblastine relies on low-yielding extraction and purification of the precursors vindoline and catharanthine from the plant Catharanthus roseus, which is then followed by simple in vitro chemical coupling and reduction to form vinblastine at an industrial scale2,3. Here, we demonstrate the de novo microbial biosynthesis of vindoline and catharanthine using a highly engineered yeast, and in vitro chemical coupling to vinblastine. The study showcases a very long biosynthetic pathway refactored into a microbial cell factory, including 30 enzymatic steps beyond the yeast native metabolites geranyl pyrophosphate and tryptophan to catharanthine and vindoline. In total, 56 genetic edits were performed, including expression of 34 heterologous genes from plants, as well as deletions, knock-downs and overexpression of ten yeast genes to improve precursor supplies towards de novo production of catharanthine and vindoline, from which semisynthesis to vinblastine occurs. As the vinblastine pathway is one of the longest MIA biosynthetic pathways, this study positions yeast as a scalable platform to produce more than 3,000 natural MIAs and a virtually infinite number of new-to-nature analogues.

Monoterpene indole alkaloids (MIAs) are a diverse family of complex plant secondary metabolites with many medicinal properties, including the essential anti-cancer therapeutics vinblastine and vincristine 1 . As MIAs are difficult to chemically synthesize, the world's supply chain for vinblastine relies on low-yielding extraction and purification of the precursors vindoline and catharanthine from the plant Catharanthus roseus, which is then followed by simple in vitro chemical coupling and reduction to form vinblastine at an industrial scale 2,3 . Here, we demonstrate the de novo microbial biosynthesis of vindoline and catharanthine using a highly engineered yeast, and in vitro chemical coupling to vinblastine. The study showcases a very long biosynthetic pathway refactored into a microbial cell factory, including 30 enzymatic steps beyond the yeast native metabolites geranyl pyrophosphate and tryptophan to catharanthine and vindoline. In total, 56 genetic edits were performed, including expression of 34 heterologous genes from plants, as well as deletions, knock-downs and overexpression of ten yeast genes to improve precursor supplies towards de novo production of catharanthine and vindoline, from which semisynthesis to vinblastine occurs. As the vinblastine pathway is one of the longest MIA biosynthetic pathways, this study positions yeast as a scalable platform to produce more than 3,000 natural MIAs and a virtually infinite number of new-to-nature analogues.
Plants produce some of the most potent human therapeutics and have been used for millennia to treat illnesses. The monoterpene indole alkaloids (MIAs) are plant secondary metabolites mostly found in Gentianales plants, bolstering a notable structural diversity and pharmaceutically valuable biological activities, with more than 3,000 MIAs derived from their common precursor strictosidine 4,5 . The MIA chemotherapeutic drugs irinotecan, vinblastine and vincristine are included in the World Health Organization list of essential medicines and exemplify some of the few established MIA therapeutics on the market, with annual market sizes of more than €1 billion (ref. 6 ). Despite the vast repertoire of MIAs with documented bioactivities, obtaining requisite amounts of MIAs needed for further development and clinical applications is often challenging because of the low yields observed in extracts from natural plant resources, as exemplified by the chemotherapeutic drugs vinblastine and vincristine requiring 500 and 2,000 kg of Catharanthus roseus dried leaves, respectively, to obtain 1 g of products 3,7,8 . Likewise, owing to the numerous stereo centres found in MIAs, multi-step synthetic chemistry holds little promise for bulk production of complex MIA therapeutics 2,9 , and thus the current vinblastine supply chains rely on extraction from C. roseus and subsequent condensation of the catharanthine and vindoline precursors 10 . As vincristine and vinblastine were listed as being drugs with a shortage in 2019-2020 by the US Food and Drug Administration 11 , there is a growing awareness of the importance of refactoring the biosynthesis of MIAs in genetically tractable heterologous hosts.
Refactoring complex plant biosynthetic pathways of human therapeutics in microbial hosts could promise more access to plant natural products and new-to-nature derivatives thereof through scalable fermentation 12 . Indeed, baker's yeast Saccharomyces cerevisiae has been engineered to produce plant-derived opiates, cannabinoids, terpenes and alkaloids to treat cancer, pain, malaria and Parkinson's disease [12][13][14][15][16] . Whereas small parts of the vinblastine biosynthetic pathway have been refactored into yeast 4,[17][18][19] , only recently was the full 31-step vinblastine biosynthetic pathway from C. roseus explained 20,21 , opening up the possibility for a demonstration of a microbial supply chain for vinblastine using engineered yeast.

A de novo strictosidine platform in yeast
The initial strictosidine strain MIA-AU was designed on the basis of a previously reported 12-step de novo strictosidine strain, including deletion of ATF1 and OYE2, encoding an alcohol acetyltransferase and a reduced nicotinamide adenine dinucleotide phosphate (NADPH) oxidoreductase, respectively, to minimize by-product formation from geraniol 4 (Fig. 1). In addition, MIA-AU was designed to down-regulate ERG20 encoding farnesyl pyrophosphate (FPP) synthase to improve the GPP:FPP ratio. However, when tested in the synthetic complete (SC) medium with all amino acids and 2% glucose, MIA-AU did not produce detectable amounts of strictosidine. Yet, when supplementing 500 μM geraniol to the medium (no higher concentration was tested owing to high toxicity of geraniol above 500 μM, Extended Data Fig. 1), MIA-AU produced 5.59 μg l −1 (10.5 nM) strictosidine, indicating geraniol precursor supply limitations (Fig. 2a). We then deleted the NADPH-dependent aldehyde reductase ARI1 and NADPH oxidoreductase OYE3, involved in syphoning 8-hydroxygeraniol and 8-oxogeranial, respectively, to shunt products 26 , resulting in strain MIA-AW-2 (Supplementary Table 2). When grown in SC medium supplemented with 500 μM geraniol, MIA-AW-2 produced 12.2 μg l −1 (23.1 nM) strictosidine (Fig. 2a). Acknowledging the challenges often encountered when expressing plant cytochrome P450s in microbes 27 , we first investigated the expression and subcellular localization of the four P450 enzymes of the strictosidine module, G8H, iridoid oxidase, 7DLH and SLS (Supplementary Fig. 1), each fused to a yeast enhanced green fluorescent protein (yEGFP). Although the fluorescence was generally low, we observed that all four P450 enzymes had correct endoplasmic reticulum membrane localization ( Supplementary Fig. 1). Using targeted proteomics, we also observed poor expression of GPP synthase, FPS and geraniol synthase, which are responsible for geraniol production, as well as iridoid oxidase and SLS (Extended Data Fig. 2b), consistent with the fluorescence microscopy result, altogether suggesting several inefficient steps requiring further optimization.
To enable iterative engineering and refactoring of long biosynthetic pathways, we included a set of unique 20-nucleotide guide-RNA sequences in front of the promoter of each expression cassette, allowing for easy swapping of target promoters or entire expression cassettes iteratively in strain MIA-BG (Methods and Extended Data Fig. 3a). Like MIA-AU, the new platform strain MIA-BG produced 9.81 μg l −1 (18.5 nM) strictosidine when fed 100 μM geraniol and tryptamine. We supplemented the cultivation medium with tryptamine in addition to one of the precursors 8-hydroxygeraniol, 8-oxogeranial, cis-trans-nepetalactol, loganic acid, loganin or secologanin to deduce possible pathway bottlenecks. Here, when feeding 8-hydroxygeraniol, MIA-BG produced 36.7 μg l −1 (69.1 nM) strictosidine, whereas feeding 8-oxogeranial or cis-trans-nepetalactol resulted in 124 μg l −1 (233 nM) or 2.51 mg l −1 (4.73 μM) strictosidine, respectively, indicating 8HGO and iridoid synthase as the main bottlenecks. Furthermore, the highest strictosidine production in MIA-BG (6.15 mg l −1 or 11.6 μM) was achieved by feeding 100 μM secologanin ( Fig. 2b and Extended Data Fig. 4). However, the unused secologanin in the spent medium further suggested STR as a rate-limiting step for strictosidine production, which could be partially alleviated by the overexpression of tCroSTR from a high copy plasmid, resulting in a substantial improvement of strictosidine titre to 39.1 mg l −1 (73.7 μM), whereas 8.53 mg l −1 (21.9 μM) secologanin remained in the medium ( Supplementary Fig. 2).
To mitigate the bottlenecks in strictosidine production, we focused first on the Pictet-Spengler-type reaction catalysed by STR by screening seven STR homologues identified from MIA-producing plants in both full-length and truncated versions (tSTR) without the N-terminal vacuolar signal peptide 4 . Individually, these variants were introduced into a strain without STR (MIA-BC), and the resulting strains were cultivated in the presence of secologanin and tryptamine. The result showed that full-length STRs yielded 2-3 times more strictosidine compared to tSTRs, with C. roseus STR (CroSTR) enabling the highest titre of 17.5 mg l −1 (32.9 μM) strictosidine (Extended Data Fig. 5a). Fluorescence microscopy analysis showed that CroSTR was localized in the vacuole or endoplasmic reticulum membrane, while the truncated variant was localized to the nucleus in fresh cultures and shifted to the cytoplasm following overnight growth ( Supplementary Fig. 3), suggesting that either the substrates secologanin and tryptamine are mostly in the vacuole, or the acidic pH in the vacuole is favourable for strictosidine synthesis. Alternatively, post-translational modifications or specific folding steps encountered along the endoplasmic reticulum-to-vacuole trafficking pathway may support STR activity. Swapping the tCroSTR in MIA-BG with CroSTR (resulting in strain MIA-BJ) enabled an efficient (80%) conversion of 50 μM secologanin to 21.2 mg l −1 (40.0 μM) strictosidine, a 6.9-fold improvement over tCroSTR (Extended Data Fig. 5b). More statistical analysis is available in the source data file.
Last, to produce strictosidine de novo from glucose and tryptophan, we increased geraniol availability by dynamically down-regulating ERG9 and ERG20 (both encoding GPP consuming enzymes) using the glucose-dependent P HXT1 and P HXT3 promoters, respectively 30 , resulting in strain MIA-CL (Supplementary Table 2). We further overexpressed a specific GPP synthase encoded by AgrGPPS2 (ref. 31 ). GgaFPS N144W , which is an FPP synthase that has lower affinity for GPP 32 and an ERG20 variant (ERG20 F96W,N127W ) that shows a low FPP synthase activity 33 . We fused ERG20 F96W,N127W to tCroGES, to enhance the GPP pool and redirect it towards the first precursor, geraniol. IDI1 and truncated HMG1 (tHMG1) were overexpressed to increase the synthesis of isopentenyl pyrophosphate and dimethylallyl pyrophosphate, both precursors for GPP 4,34 . Finally, CroTDC was overexpressed to convert tryptophan to tryptamine, resulting in strain MIA-CM-5 (Supplementary Table 2). When grown in SC medium, strain MIA-CM-5 produced 7.43 mg l −1 (19.0 μM) loganin and 13.3 mg l −1 (25.0 μM) strictosidine (Fig. 2c). As glucose is depleted in 1-2 days during the 6-day cultivation, and yeast continues to use ethanol produced from glucose, we replaced the constitutively active P TEF2 and P CCW12 promoters driving expression of genes encoding AgrGPPS2 and ERG20 F96W,N127W -tCroGES fusion with glucose-repressible promoters P MLS1 and P ICL1 to avoid competition for GPP during growth on glucose and to ensure high expression on ethanol after glucose was depleted. In SC medium, MIA-CM-3 produced 15.0 mg l −1 (38.6 μM) loganin and 11.4 mg l −1 (21.5 μM) strictosidine, concentrations similar to those produced by MIA-CM-5 (Fig. 2c). However, when grown in yeast extract peptone dextrose (YPD) medium, MIA-CM-3 reached 25.2 mg l −1 (47.4 μM) strictosidine, which was considerably higher than MIA-CM-5 (8.73 mg l −1 or 16.5 μM), and both MIA-CM-5 and MIA-CM-3 strains produced notably lower loganin in YPD medium compared to SC medium (Fig. 2c-d and Supplementary Fig. 4).

Engineering of the SGD gateway
Having optimized de novo strictosidine production, we next sought to express the gateway enzyme strictosidine-β-d-glucosidase (SGD), together with one-step conversion of strictosidine aglycone to tetrahydroalstonine (THA), which is not part of the main vinblastine pathway but is the shortest downstream pathway to a side product for which a chemical standard is available 35 (Extended Data Fig. 8a). Despite extensive studies on SGD, including a crystal structure of the Rauvolfia serpentina SGD (RseSGD) and functional expression of CroSGD in yeast 36 , we initially did not detect any strictosidine aglycone when expressing CroSGD alone, or THA when expressing CroSGD together with C. roseus THA synthase (CroTHAS) in strain MIA-BJ ( Fig. 3a and Supplementary Fig. 5). Several attempts to express full-length or truncated CroTHAS (removing the nuclear localization sequence K 214 K 215 K 216 R 217 ) fused to full-length or truncated CroSGD (removing the C-terminal nuclear localization sequence K 537 KRFREEDKLVELVKKQKY 555 ) did not yield detectable THA production in yeast (data not shown). However, when testing 46 homologues of CroSGD, RseSGD yielded 4.02 mg l −1 (11.4 μM) THA from feeding 100 μM secologanin and 1 mM tryptamine, and another 15 homologues yielded more than 0.1 μM or 35.2 μg l −1 (0.1% yield from 100 μM secologanin feeding) THA (Extended Data Fig. 8b). Despite CroSGD being previously reported to localize to the nucleus in plant cells 35 , fluorescence microscopy of yeast cells expressing N-terminally yEGFP-tagged CroSGD and RseSGD showed different subcellular localization, with RseSGD localizing in the nucleus and CroSGD showing diffuse cytoplasmic distribution ( Supplementary  Fig. 6). As SGD is reported to form a protein complex in the nucleus in plant cells 35 , the cytoplasmic location of CroSGD was presumably the main reason for the lack of activity from CroSGD. The distinct protein localization/activity was surprising given the 70% protein identity between these two homologues. To investigate sequence or structural features indispensable for RseSGD function in yeast, we divided SGD into four domains on the basis of sequence conservation between RseSGD and CroSGD ( Fig. 3c and Supplementary Table 3), and tested THA production from all possible 16 hybrid SGDs constructed from the four domains of the two SGDs. From this, we observed that the third domain from RseSGD (R 3 ), is required for SGD activity while maintaining the integrity of the third and fourth domains (R 3 +R 4 ) produced the highest amount of THA (4.69 mg l −1 or 13.3 μM) (Fig. 3c), probably contributing to SGD complex formation or correct localization in the nucleus ( Supplementary Fig. 6).

Functional expression of MIA modules
Next, to refactor the tabersonine/catharanthine module from C. roseus in yeast, we first constructed strain MIA-DC by integrating all ten genes from STR (to enable strictosidine production from fed secologanin and tryptamine) to tabersonine and catharanthine synthases together with CroCPR and CroCYB5 for the cytochrome P450 enzyme geissoschizine oxidase 20,21 ( Fig. 1 and Extended Data Fig. 9). When fused to a C-terminally tagged yEGFP, all proteins, except for geissoschizine synthase and PAS, showed relatively strong fluorescence (Supplementary Fig. 7). When fed 100 μM secologanin and 1 mM tryptamine, strain MIA-DC produced 9.29 μg l −1 (27.6 nM) tabersonine and 6.96 μg l −1 (20.7 nM) catharanthine (Extended Data Fig. 9). In parallel, we refactored the vinblastine module (strain MIA-DE) by integrating all nine genes from tabersonine 16-hydroxylase (encoded by T16H1 and T16H2) to peroxidase PRX1, which was suggested to condense vindoline and catharanthine, together with CroCPR and CroCYB5, into a wild-type strain 17,19 (Fig. 1 and Extended Data Fig. 9). When fed 100 μM tabersonine and 100 μM catharanthine, strain MIA-DE produced 0.992 mg l −1 (2.17 μM) vindoline but no detectable vinblastine (Extended Data Fig. 9). Targeted metabolite analysis identified a strong signal for masses identical to 16-hydroxytabersonine and 16-methoxytabersonine, indicating a potential bottleneck at the P450 enzyme T3O (Extended Data Fig. 9). This was consistent with the improved vindoline production when extra copies of two consecutive genes in the vinblastine module were expressed in the MIA-DE strain, albeit with overexpression of T16H2 and 16OMT yielding the highest improvement (more than twofold) in vindoline titre compared to strain MIA-DE (Extended Data Fig. 9). Whereas strain MIA-DE did not produce vinblastine (C 46 H 58 N 4 O 9 ) from vindoline and catharanthine feeding, high-resolution MS detected a compound with formula C 46 H 56 N 4 O 9 in both wild-type and MIA-DE strains, suggesting non-enzymatic leurosine formation as also previously shown to be synthesized in vitro from catharanthine and vindoline 37,38 . Confocal microscopy analysis indicated that CroPRX1 was not expressed/folded properly ( Supplementary Fig. 8), and several attempts to truncate PRX1 at the N and/or C termini did not yield any vinblastine production when cells were grown in media supplemented with 100 μM catharanthine and 100 μM vindoline. Integrating the vinblastine module into strain MIA-DC (catharanthine-tabersonine strain) yielded strain MIA-DJ, which produced 44.0 μg l −1 (130 nM) catharanthine and 4.37 μg l −1 (9.57 nM) vindoline when cultivated in YPD medium supplemented with 100 μM secologanin and 1 mM tryptamine, whereas strain MIA-DC produced 11.4 μg l −1 (34.0 nM) tabersonine and 90.8 μg l −1 (270 nM) catharanthine, indicative of catharanthine synthase having better expression or higher catalytic reactivity than tabersonine synthase (Extended Data Fig. 9). These results are in line with untargeted proteomics data showing strain MIA-DJ had relatively low expression of tabersonine synthase (Extended Data Fig. 10).
To enable one-pot de novo production of vindoline and catharanthine, C. roseus catharanthine synthase was integrated into strain MIA-CW-1 under the control of a galactose inducible (P GAL1 ) promoter. Extra copies of the ERG20**-tCroGES fusion and AgrGPPS2 were integrated to boost geraniol synthesis. To enhance the conversion of the unstable and highly reactive intermediate strictosidine aglycone 23 to geissoschizine and its oxidation, we also integrated an extra copy of CroGS, CroGO and CroCPR, resulting in the strain MIA-EM-1. Using biphasic fermentation with glucose and galactose as carbon sources, MIA-EM-1 produced 2.32 μg l −1 (5.08 nM) vindoline and 26.6 μg l −1 (79.1 nM) catharanthine de novo ( Fig. 4 and Extended Data Fig. 10). We further constructed strain MIA-EM-2 by integrating an extra copy of each of the six tabersonine module genes, namely CroGS, CroGO, CroSAT, CroPAS, CroDPAS and CroTS, into strain MIA-EM-1. Similarly, we constructed MIA-EM-3 by integrating an extra copy of CroTS as well as the vindoline module genes (CroT16H2, CroT3O, CroT3R, CroD4H and CroDAT) (Extended Data Fig. 10). Under the same fed-batch cultivation condition, strain MIA-EM-2, which had an enhanced tabersonine module, produced 5.48 μg l −1 (12.0 nM) vindoline and 69.4 μg l −1 (206 nM) catharanthine (Extended Data Fig. 10). On the other hand, strain MIA-EM-3 (with a further copy of vindoline module genes) produced 4.19 μg l −1 (9.18 nM) vindoline and only one out of three colonies produced 6.46 μg l −1 (19.2 nM) catharanthine (Extended Data Fig. 10), indicating that the main bottleneck was still within the tabersonine/ catharanthine module and further enhancing the vindoline module expression was detrimental to catharanthine production because of a stronger pull of dihydroprecondylocarpine acetate, which is also a precursor to catharanthine (Fig. 1). Subsequently, strain MIA-EM-2 was grown in fed-batch cultivation in ambr 250 bioreactors, with continuous feeding of 3× SC containing 2% glucose and 3 mM tryptophan (between 20 and 72 h), and daily dosing of 3 g l −1 galactose during days 4-10. Samples were collected every 24 h to analyse the MIA metabolites. At the end of the fermentation (day 11), the catharanthine and vindoline titres reached 91.4 μg l −1 (271 nM) and 13.2 μg l −1 (29.0 nM), respectively (Extended Data Fig. 11 and Supplementary Fig. 9).

Vinblastine semisynthesis
Vindoline and catharanthine can be enzymatically coupled using a peroxidase and peroxide as demonstrated previously 25 , chemically coupled using Fe(III) 2 or photo-chemically coupled using near-UV radiation in the presence of flavin mononucleotide 38 . The enzymatic coupling using horseradish peroxidase did not yield a detectable amount of vinblastine (data not shown), and we could only produce 336 μg l −1 (414 nM) and 397 μg l −1 (490 nM) vinblastine from pure catharanthine (100 μM) and vindoline (10 μM) standards using the Fe(III)-based coupling and photo-chemical methods, respectively (Extended Data Fig. 12). Also, attempts to couple yeast-produced catharanthine and vindoline (after liquid-liquid extraction and roughly 750-fold concentration of the compounds, see Supplementary Methods) using the Fe(III) method was unsuccessful (data not shown), presumably because of unspecific coupling to by-products or a strong suppression of vinblastine signal by the complex matrix concentrated from the fermentation broth. However, following purification using thin-layer chromatography and subsequent preparative high-performance liquid chromatography (HPLC) (Supplementary Methods), we could detect 23.9 μg l −1 (29.4 nM) vinblastine using the Fe(III) coupling method (Extended Data Fig. 12).
A considerable amount of the intermediate anhydrovinblastine was also formed yet was not quantifiable owing to a lack of an analytical standard.

Discussion
Here we report the optimization of the strictosidine biosynthetic pathway by more than 1,000-fold, reaching titres of 25.2 mg l −1 (47.4 μM) in 96-well plates (Fig. 2d). From this, we demonstrated de novo production of catharanthine and vindoline, with the 30-step vindoline pathway representing a very long heterologous pathway refactored in a microbial cell. Furthermore, and in alignment with industrial scale manufacturing of vinblastine, catharanthine and vindoline were purified from fermentation broths and chemically coupled to form vinblastine (Extended Data Fig. 12). Although we were unable to make vinblastine directly in yeast (as CroPRX1 proved difficult to express in yeast as indicated by proteomics analysis of the MIA-DJ strain), our microbial process mimics the plant-based production process, which relies on coupling vindoline and catharanthine after their purification from C. roseus leaves. Although it may one day be possible to identify a PRX1 homologue that successfully catalyses the reaction when compartmentalized in the peroxisome, nevertheless this study provides a demonstration of a microbial supply chain for a complex essential anti-cancer medicine. Microbial supply chains for essential medicines with few other suppliers, long lead times and high manufacturing complexity, are viable solutions to secure the global or regional supply of these drugs, especially during unpredictable environmental events, such as natural disaster, plant disease and pandemics 39 . This said, further metabolic engineering and bioprocess optimization are needed for engineered yeast to supplement and ultimately stabilize existing C. roseus-based vinblastine supply chains. Specifically, the tabersonine/catharanthine module, which had many promiscuous reactions, potential host-heterologous pathway interactions and protein aggregation, was particularly challenging 20 . Furthermore, expression of the seven P450 enzymes could be further optimized, or they could be replaced with better homologues. Likewise, de novo vindoline production required more deletion, knockdown and overexpression of ten native yeast genes to improve precursor supplies, which also need to be regulated to ensure balanced cell growth and production in future process optimizations. Having said this, we consider both the strictosidine platform, and the demonstration of one-pot de novo vindoline and catharanthine production ripe for new MIA pathway discoveries as well as production and bioactivity testing of new unnatural MIA analogues with higher potency and/or efficacy, such as 10-fluoro-vinblastine 40 , in yeast cell factories.

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-022-05157-3.

Chemical standards
All chemical standards (Supplementary Table 2) had a purity of 95% or higher.

Gene discovery
Sequence for Vmi8HGO-A was obtained from V. minor transcriptome data 41 using BLASTn 42 and the C. roseus 8HGO-A sequence (GenBank accession no. AHK60836.1) as the query. Open reading frames were predicted using TransDecoder (http://transdecoder.github.io).

Plasmid construction
All plasmids were constructed by USER cloning 43 or restriction enzyme cloning, followed by chemical transformation into E. coli DH5α competent cells, and plated on Luria-Bertani agar containing 100 μg ml −1 ampicillin and grown at 37 °C overnight (16-20 h). E. coli transformants were genotyped by colony PCR, and colonies with correct size of insertion were grown in 5 ml of Luria-Bertani medium containing 100 μg ml −1 ampicillin at 37 °C for 16-20 h, followed by miniprep plasmid extraction and validated by Sanger sequencing. All biosynthetic genes (Supplementary Table 1) were codon optimized for yeast expression and synthesized by either IDT or Twist Biosciences. All genes were assembled as expression cassettes in either centromeric (pRS series) or 2-μm (pESC series) plasmids by USER cloning. Single guide-RNA plasmids were constructed by USER cloning a double-stranded oligo containing the 20-nucleotide guide-RNA sequence into a 2-μm vector with either the LEU2 or URA3 selection marker. Several guide-RNA plasmids were constructed by USER assembly of all guide-RNA expression cassettes amplified from single guide-RNA plasmids. All plasmids (Supplementary Table 4) were verified by Sanger sequencing of the insertion.

Yeast strain construction
All yeast strains (Supplementary Table 2) were constructed on the basis of the wild-type strain CEN.PK2-1C using the lithium acetate/ single-stranded carrier DNA/PEG protocol 44 , and have the same auxotrophy for uracil, histidine, leucine and tryptophan. Genome editing (integration, deletion or swapping) was performed using the CasEMBLR method 45 . All DNA parts were amplified from precloned plasmids using Phusion High-Fidelity PCR Master Mix with HF Buffer (Thermo Fisher Scientific) using primers containing 25-30 bp overhangs (IDT). For strains MIA-AU and MIA-AW-2, SpyCas9 was expressed from a centromeric plasmid with the TRP1 selection marker. For all other MIA strains, CEN.PK2-1C wild-type strain was first transformed with a plasmid (TRP1) harbouring the SpyCas9 cassette, followed by another transformation to integrate the SpyCas9 cassette into EasyClone site XII-1 for stable expression 46 . For all editing were transformed with a plasmid expressing a single or multiple-targeting guide-RNA fragments, as well as DNA parts as repair templates Transformants were plated on SC agar lacking the appropriate amino acid(s) (uracil, leucine, histidine or tryptophan) for selection of the guide-RNA plasmid, and grown for 3-4 days at 30 °C. All strains with genomic integration were confirmed by genotyping PCR and Sanger sequencing of the integrated gene cassettes.
For all strictosidine strains except MIA-AU and MIA-AW-2, a unique guide-RNA sequence was included at the beginning of all gene cassettes (beginning of the promoter sequence) for easy swapping of the promoter or deletion of whole gene cassettes (Extended Data Fig. 3a).
To swap a promoter/gene in an expression cassette, yeast cells were transformed with a plasmid expressing the guide-RNA, together with a new promoter/gene and parts overlapping with sequences up-and downstream of this promoter/gene (Extended Data Fig. 3b). Similarly, to delete a whole cassette, yeast cells were transformed with a plasmid expressing the specific guide-RNA and two parts (with a 50-60 bp overlapping sequence) that are homologous to the up-and downstream of the locus to be deleted (Extended Data Fig. 3c). Transformants were plated on SC agar lacking either leucine or uracil for selection of the guide-RNA plasmid and confirmed by colony PCR and sequencing.

Media and yeast cultivation
Yeast cells were cultivated in either SC defined medium or YPD rich medium, both containing 2% glucose as the carbon source (except for the screening of the iridoid synthase + cyclase and 8HGO variants, which used glucose and galactose). For batch-fed cultivations, 3× SC medium (2% glucose) supplemented with 3 mM tryptophan was used to reach a high biomass.
For all small-scale cultivations, precultures were typically made by growing yeast colonies (biological replicates) in 150 μl of SC medium (lacking appropriate amino acid for auxotrophic selection) or YPD medium on a 96-well microtitre plate and incubated at 30 °C shaking at 300 r.p.m. overnight (roughly 16-20 h). The maximum specific growth rate for all yeast strains were characterized using the Growth Profiler 960 system (Enzyscreen) (Supplementary Fig. 9). All cultures were grown in 300 μl of SC medium in 96-well square-shaped microtitre plates for at least 3 days. Cultures were grown at 30 °C with shaking speed of 250 r.p.m. and the cell density was measured every 30 min. Data from the exponential growth phase were used to determine the maximum specific growth rate by linear regression. Small-scale production assays were performed using either 96-well (with 0.5 ml of medium) or 24-well (with 2 ml of medium) deep-well plates. After 1 day (typically reaching the stationary phase), 10 μl of precultures were transferred to either SC lacking appropriate amino acid(s) or YPD medium supplemented with one or more precursors. For testing strictosidine production, yeast strains were fed with 0.1-1 mM tryptamine and one of the precursors including geraniol, 8-hydroygeraniol, 8-oxogeranial, cis-trans-nepetalactol, loganic acid, loganin and secologanin. For testing tabersonine and catharanthine production, cells were fed with 1 mM tryptamine and 50-250 μM secologanin. For testing vindoline or vinblastine production, cells were fed with 100 μM tabersonine or 100 μM vindoline and 100 μM catharanthine, respectively. All compounds used in the feeding experiments are listed in Supplementary Table 5.
Small-scale fed-batch cultivations were performed in BioLector Pro systems using flower-shaped plates. The fermentation started with inoculating 1 ml 3× SC (2% glucose) medium supplemented with 3 mM tryptophan to an initial OD 600 of 0.5. Continuous exponential feeding with the same medium except containing 36% glucose started at 20 h using the following equation: where t is the feeding time (h). Temperature was kept at 30 °C throughout the whole cultivation. The pH was controlled at 5.5 using 10% NH 4 OH solution. The plate was shaken at speed 1,000 r.p.m. The relative humidity in the growth chamber was maintained at 85% using distilled water to minimize evaporation of the media. An automated liquid handler was used to take 0.1 ml culture samples on days 3, 5 and 7 for metabolite analysis.
Batch-fed fermentations were performed on an ambr 250 system using single-use microbial vessels. The fermentation started by inoculating 100 ml of 3× SC medium (2% glucose) supplemented with 3 mM tryptophan to an initial OD 600 of 5.0. The same medium, except containing 36% glucose was fed continuously into the bioreactor between 20 and 96 h using the following equation: Flow rate (ml h ) = 48e t −1 0.0125 Between days 4 and 10, 3 g l −1 of galactose was added into the fermentation every day to induce the expression of catharanthine synthase. The temperature was kept at 30 °C throughout the whole fermentation. The pH of all bioreactors was controlled at 5.0 using 10% NH 4 OH solution. The air flow rate was initially at 1 volume of air per unit of medium per unit of time (vvm), and dissolved oxygen (dO 2 ) was maintained above 40% by increasing the agitation speed. CO 2 in the exhaust gas was monitored by a gas analyser. An automated liquid handler was used to take 1.0 ml broth samples every 24 h for metabolite analysis. At the end of the cultivation, all (150-200 ml) broth was used for crude extraction of MIA compounds including catharanthine and vindoline.

MIA compounds extraction and purification
In some cases, MIA metabolites were extracted from fermentation broths using the Oasis HLB (hydrophilic-lipophilic balance) cartridges and 96-well plates following the manufacturer's manual (Waters). Solid phase extraction (SPE) of all samples (0.5-2 ml) from 96-or 24-well deep plates were performed on Oasis HLB 96-Well Plates (SKU 186000679) and eluted with 0.5 ml of pure methanol. Samples taken from the ambr 250 bioreactors were extracted using dichloromethane and the resulting crude extract was further purified by preparative thin-layer chromatography followed by preparative HPLC (Supplementary Methods). The final samples were dissolved with appropriate amounts of water and used for chemical coupling to produce vinblastine.

Synthesis of vinblastine from vindoline and catharanthine
Photo-chemical coupling of vindoline and catharanthine was performed in 0.1 M Tris-HCl buffer pH 7.0 containing 1 mM MnCl 2 (ref. 38

Metabolites sample preparation and analysis
After 3 or 6 days, 200 μl of cultures were mixed with 20 μl of 10 mg l −1 caffeine solution (as the internal standard for normalization) and filtered through a filter plate (PALL, AcroPrep Advance, 0.2 μm Supor membrane for media/water) by centrifugation at 2,200g for 1 min. For tabersonine and vindoline analysis, strains MIA-CR-A and MIA-CW-1 were also cultivated in 4 ml of YPD medium on a 24-well deep-well plate for 6 days, and the whole spent medium was used for SPE using Oasis HLB 96-well plate (Waters) following manufacturer's protocol. Metabolites were eluted with 0.8 ml of pure methanol, evaporated to dryness at room temperature and reconstituted in 100 μl of water and mixed with 10 μl of 10 mg l −1 caffeine solution before filtration. A series of standard mixtures (Supplementary Table 5) were prepared in the same way as analytical samples. Typically, 1 μl samples were injected for analysis. For de novo tabersonine, catharanthine and vindoline strains, 3 μl samples were injected to produce a higher signal. Targeted metabolite analysis of loganic acid, loganin, secologanin, strictosidine, tabersonine, catharanthine, vindoline and vinblastine was conducted using an Advance ultra-HPLC (UHPLC) system (Bruker Daltonics) coupled to an EVOQ Elite triple quadrupole mass spectrometer (Supplementary For the analysis of non-polar metabolites (geraniol and 8hydroxygeraniol, and by-product citronellol), cells were grown in 2 ml of medium using 24-well deep plates. Geraniol and 8-hydroxygeraniol were extracted by mixing 1.6 ml samples with 400 μl of ethyl acetate (containing 50 μg ml −1 nerol as the internal standard), shaken at 4 °C for 30 min. The aqueous and organic phases were separated by centrifugation at 13,000g for 1 min, and 250 μl of the upper (ethyl acetate) phase was transferred to a gas chromatography vial with glass insert and analysed on a gas chromatography-flame ionization detector. Details for the gas chromatography and LC-MS programmes are described in the Supplementary Methods.

Fluorescence microscopy
The yEGFP was fused to the N or C terminus of target proteins depending on their reported or predicted structural information. All yEGFP proteins were integrated into the wild-type yeast strains and confirmed with colony PCR. Localization markers for nucleus (Addgene plasmid no. 133648), endoplasmic reticulum membrane (Addgene plasmid no. 133647), mitochondria (Addgene plasmid no. 133655) and vacuole (Addgene plasmid no. 133654), containing mCherry or DuDre, were ordered from Addgene and integrated into the strains with yEGFP fusion protein as described previously 47 . Strains were cultivated in SC-Trp medium overnight (16-20 h) and imaged in the morning, or overnight culture was diluted 1:5 and cultivated for 4 h before analysis. Then 3 μl of culture was applied to an objective glass, covered with a cover glass and imaged immediately. Confocal images were acquired with a laser-scanning upright confocal microscope LSM 700, Axio Imager 2 (Carl Zeiss, Inc.), equipped with four diode lasers (405, 488, 555, 639 nm), using a C-Apochromat ×63/1.2 W Korr M27 water objective and ×4 scan zoom. Pinhole size was set to 1 Airy unit. Samples were illuminated with 488 and 555 nm lasers at 0.4-4 mW. 512 × 512 pixel images were acquired using photomultiplier tube detectors. yeGFP acquisition parameters were: 300-578 nm (emission wavelength range), 0.39 μs (line time), 600-1,000 (gain) and 0 (offset). mCherry/DuDre acquisition parameters were: 578-800 nm (emission wavelength range), 0.39 μs (line time), 600-1.000 (gain) and 0 (offset). A line average of eight was applied to both channels. Images were processed with ZEN v.3.2 (blue edition) (Carl Zeiss Microscopy GmbH).

Analysis of proteins
All yeast strains were grown in 2 ml of SC or YPD media in three biological replicates at 30 °C 300 r.p.m. for 2 days. Cells (roughly 50 OD × ml units) were pelleted by centrifugation at 5,000g for 3 min and kept at −80 °C until lysis and sample preparation. Yeast cells were lysed and protein extracted as described previously 48 . Briefly, it consists of a bead-beating step, cell lysis, protein precipitation, protein resuspension, protein quantification and normalization of protein concentration followed by standard bottom-up proteomic procedures of reducing and blocking cysteine residues and tryptic digestion. Peptide samples were analysed on an Agilent 1290 UHPLC system coupled to an Agilent 6460QQQ MS (Agilent Technologies) as described previously 49 . Briefly, peptide samples were loaded onto an Ascentis ES-C18 Column (Sigma-Aldrich) and introduced to the MS using a Jet Stream source (Agilent Technologies) operating in positive-ion mode. The data were acquired with Agilent MassHunter Workstation Software v.B.08.02. LC-MS Data Acquisition operating in dynamicMRM mode. Multiple reaction monitoring (MRM) transitions for the targeted proteins were generated by Skyline software v.20.2 (MacCoss Laboratory Software) and selection criteria excluded peptides with Met/Cys residues, tryptic peptides followed by further cut sites (KK/RR) and peptides with proline adjacent to K/R cut sites. The data and Skyline methods are available on Panoramaweb (https://panoramaweb.org/microbial-synthesis-of-vinblastine.url). Proteins from the MIA-DJ (tabersonine-vinblastine double module) strain were analysed using the Cap-LC system equipped with a C18 easy spray column (Thermo Fisher Scientific), coupled to Orbitrap Q Exactive HF-X mass spectrometer (Thermo Fisher Scientific) 50 . The resulting data were analysed using Proteome Discover v.2.3 (Thermo Fisher Scientific) by searching against the S. cerevisiae proteome data (Uniprot ID UP000002311) combined with all heterologous protein sequences. The abundance of each protein is reported as the relative intensity to the total intensity of all identified peptides.

Reporting summary
Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
All metabolite data shown in figures and extended data figures are available in the Source data provided with this paper. Accession of all heterologous genes used in this study are listed in Supplementary  Table 1. Targeted proteomics data for strain MIA-AU are available at ProteomeXchange with identifier PXD024976 (https://panoramaweb. org/microbial-synthesis-of-vinblastine.url). Untargeted proteomics data for strain MIA-DJ are available at ProteomeXchange with identifier PXD025067.