Mechanism-based tuning of insect 3,4-dihydroxyphenylacetaldehyde synthase for synthetic bioproduction of benzylisoquinoline alkaloids

Previous studies have utilized monoamine oxidase (MAO) and L-3,4-dihydroxyphenylalanine decarboxylase (DDC) for microbe-based production of tetrahydropapaveroline (THP), a benzylisoquinoline alkaloid (BIA) precursor to opioid analgesics. In the current study, a phylogenetically distinct Bombyx mori 3,4-dihydroxyphenylacetaldehyde synthase (DHPAAS) is identified to bypass MAO and DDC for direct production of 3,4-dihydroxyphenylacetaldehyde (DHPAA) from L-3,4-dihydroxyphenylalanine (L-DOPA). Structure-based enzyme engineering of DHPAAS results in bifunctional switching between aldehyde synthase and decarboxylase activities. Output of dopamine and DHPAA products is fine-tuned by engineered DHPAAS variants with Phe79Tyr, Tyr80Phe and Asn192His catalytic substitutions. Balance of dopamine and DHPAA products enables improved THP biosynthesis via a symmetrical pathway in Escherichia coli. Rationally engineered insect DHPAAS produces (R,S)-THP in a single enzyme system directly from L-DOPA both in vitro and in vivo, at higher yields than that of the wild-type enzyme. However, DHPAAS-mediated downstream BIA production requires further improvement.

R ecent progress in synthetic biology and metabolic engineering offers potential to optimize the bioproduction of virtually any compound [1][2][3][4][5] . Engineered metabolic systems are now being designed for production of biofuels, polymers, bulk chemicals, and pharmaceuticals 1,2 . Computational methods to select the best metabolic pathways from thousands of possibilities have been demonstrated for production of 1,4-butanediol 4 . The combination of metabolic engineering with enzyme engineering can even be used to construct artificial biosynthetic pathways to compounds including 2,4-dihydroxybutyric acid 5 .
With the goal of further developing enzyme engineering as a practical approach that can expand the scope of bioproduction targets, benzylisoquinoline alkaloid (BIA) biosynthesis is presented as an example for synthetic enzyme development. Opioid analgesics are derived from BIAs and are essential medicines for pain and palliative care as defined by the World Health Organization 6 . Microbial production of pharmaceutical alkaloids offers advantages in terms of cost efficiency, environmental sustainability, and process control 7 . Recently, bioproduction of opioid alkaloids has been achieved in yeast via a norcoclaurine (higenamine) containing pathway 8,9 and in Escherichia coli via a tetrahydropapaveroline (THP, norlaudanosoline) containing pathway 10,11 . The THP pathway requires one less enzymatic step and has afforded the highest BIA titer of 1 mM THP (287 mg/L). However, current THP bioproduction relies upon monoamine oxidase (MAO), a membrane bound flavoenzyme that is active towards many monoamines in addition to dopamine 12,13 . It is therefore desirable to engineer a soluble enzyme to improve production of THP, and downstream alkaloids including thebaine, which has been recently reported with microbial titers lower than 10 mg per litre 11 .
Utilizing BIA bioproduction as a model pathway, the current system aims to search for alternative enzyme engineering targets and learn alternative biosynthetic pathways. This synthetic biology workflow includes enzyme selection and learning for synthetic pathway design via the recently developed Mpath prediction software 14 . Specifically, a 3,4-dihydroxyphenylacetaldehyde synthase (DHPAAS) is identified from Bombyx mori and engineered to switch between two distinct activities for the direct production of THP in a symmetrical pathway. Existing reports have applied substrate specificity engineering to metabolic engineering 5 . In contrast, the current study applies functional enzyme engineering to the assembly of an alternative bioproduction pathway.

Results
Pathway design and enzyme selection. In 2014, the M-path computational platform was developed to predict putative metabolic pathways and enzymes that might catalyze new reactions 14 . M-path uses an iterative random algorithm to score chemical similarities and can be operated as a web-based tool. In contrast to searching known enzyme networks, M-path is advantageous in that it can predict unknown enzymatic reactions based upon substrate and product similarities. Furthermore, M-path can find reactions and pathways from a wide range of search space and easily expand the search space 14 . To explore alternative BIA production pathways from aromatic amino acids, the M-path search algorithm was tested. A combined database of updated enzyme entries from BRENDA (BRaunschweig ENzyme DAtabase) 15 and Kyoto Encyclopedia of Genes and Genomes (KEGG) 16 was used to increase enzyme targets. Aromatic aldehyde synthase (AAS) and DHPAAS were identified as putative shortcuts for production of 4-hydroxyphenylacetaldehyde (4-HPAA or 4-HPA) from L-tyrosine (Tyr), and 3,4-dihydroxyphenylacetaldehyde (DHPAA, DHPA or DOPAL) from L-3,4dihydroxyphenylalanine (L-DOPA) (Fig. 1, Supplementary Table 1). Although authors were aware of the function of DHPAAS, this  Table 1). b Multiple pathways lead to THP and norcoclaurine production, including a nonsymmetrical DDC-MAO-mediated pathway (blue and grey arrows), symmetrical DDC-DHPAAS/AAS-mediated pathways (red and grey arrows), and an engineered DHPAAS single enzyme system (green split arrows). (S)-norcoclaurine can be produced from 4-hydroxyphenylacetaldehyde (4-HPAA) and dopamine via the enzyme (S)norcoclaurine synthase (NCS, EC 4.2.1.78). Dopamine and DHPAA undergo spontaneous Pictet-Spengler condensation to form THP (3-hydroxynorcoclaurine), or this reaction can also be catalyzed by NCS. THP is converted to reticuline via norcoclaurine 6-O-methyltransferase (6-OMT), coclaurine N-methyltransferase (CNMT) and 3-hydroxy-N-methylcoclaurine 4-O-methyltransferase (4-OMT). An additional enzyme, N-methylcoclaurine 3hydroxylase (NMCH), is necessary to produce reticuline from norcoclaurine. The 4-HPAA containing pathways are shown in grey as the current study focuses on the DHPAA containing pathway. All structures were drawn with ChemDraw example of enzyme selection illustrates the importance of updating enzyme databases for prediction of recently characterized enzymes, as new functions are continuously discovered from nature and enzyme engineering 17  The construction of symmetrical AAS-and DHPAAS-mediated BIA production pathways may offer some advantages over the corresponding nonsymmetrical MAO-mediated pathways (Fig. 1). This includes increased specificity of soluble DHPAAS for L-DOPA relative to the specificity of MAO for dopamine. Mathematical models and numerical simulations were employed to compare the production of THP via the nonsymmetrical and symmetrical pathways ( Supplementary Figures 1-3, Supplementary Table 2). Monte-Carlo simulation was performed with randomly generated values for parameters lacking reported data as described in Supplementary Methods and Supplementary Table 2. Later in vitro and in vivo tests suggested that highly reactive DHPAA may be depleted by chemical or enzymatic side-reactions. Including the carbon loss of DHPAA as 'drain' in dynamic models resulted in slightly lower THP yields (Fig. 2), indicating a better match to experimental levels. However, many diverse variables, including growth medium composition, pH, temperature, potential inhibitors, cofactor recycling, genetic regulation, and metabolic flux, should also be considered as learning data for the improvement of THP production. When factoring in product feedback inhibition 21 together with DHPAA drain, symmetrical DDC-DHPAAS pathways produced higher predicted yields of THP than those of MAO-mediated pathways (Fig. 2, Supplementary Figure 3). These models suggest that DHPAAS-mediated pathways might offer clues to produce THP at levels higher than the best reported MAO-mediated THP benchmark of 1 mM (287 mg/L) 10 . Moreover, the improved performance of the feedback inhibition models emphasizes that the balance of dopamine and DHPAA is critical for optimal THP production. Therefore tuning the production of both DHPAA and dopamine by DHPAAS was pursued further.
Enzyme design and structure-based engineering of DHPAAS. Comparison of putative plant AAS and insect DHPAAS structures was performed to rationally select optimal enzyme sequences for BIA production. Dimeric homology models of putative AAS and DHPAAS in complex with aromatic amino acid substrate covalently linked to PLP cofactor were generated THP Yield (%)  with MODELLER 22 operated within Chimera 23 using default settings, followed by structure refinement with MOE ( Fig. 3).
M-path identified 4-HPAAS as an enzyme to produce 4-HPAA, a key intermediate in plant BIA synthesis. Accordingly, we hypothesized that P. somniferum might utilize AAS activity for natural 4-HPAA bioproduction and explored P. somniferum sequences as potential AAS enzymes. Interestingly, P. somniferum tyrosine decarboxylase 1 (TyDC1), which was modeled based on the structure of Sus Scrofa DDC in complex with carbidopa (PDB ID: 1JS3) 24 , contains a leucine residue at the position corresponding to AAAD active site His192 (Fig. 3d, e), bringing attention to the 192 position as an important catalytic residue. However, aside from the unique TyDC1 Leu205, all P. somniferum TyDC sequences highly resemble that of canonical AAAD. In contrast, additional active site differences are observed when comparing putative insect DHPAAS sequences (Fig. 3). Therefore, focus was shifted towards insect DHPAAS for the selection of an optimal BIA bioproduction enzyme.
Many questions still remain about the evolution of insect DHPAAS, and its oxidative decarboxylation mechanism including the elucidation of all essential catalytic residues. In order to help clarify these questions and gain insight into mechanismbased engineering of DHPAAS, phylogenetic classification was carried out in combination with structural analysis. Phylogenetic analysis of 738 insect AAAD-related sequences resulted in the identification of 247 putative DHPAAS sequences and five distinct DHPAAS groups (Fig. 4, Supplementary Table 3).
Uncharacterized Lepidoptera DHPAAS comprises the central phylogenetic group (Fig. 4), and was selected as a candidate for DHPAAS mechanism studies. When analyzing insect DHPAAS, the unique region formed by residues Gly353 to Arg324 could not be well modeled using the structure of D. melanogaster DDC (PDB ID: 3K40) 25 as the template. The structure of this 320-350loop region, which participates in cross-dimer active site formation and substrate binding, improved when using a template of human histidine decarboxylase in complex with histidine methyl ester (PDB ID: 4E1O) 26 .
Comparison of DDC and DHPAAS active sites indicates that residue 192 (B. mori and D. melanogaster DHPAAS numbering) determines catalytic activity as that of a decarboxylase or aldehyde synthase (Fig. 3 and Supplementary Table 3). This 192 residue can form a hydrogen bond with the PLP-aromatic amino acid external aldimine, which undergoes oxidation in the DHPAAS mechanism. Aedes aegypti and D. melanogaster DHPAAS have been reported with Asn192 residues 20,27 . However, during the course of the current study, Asn192 was independently identified and verified as the key catalytic residue via structural and functional analysis.  Table 3). Tyr79-Phe80 is conserved in insect DDC, however this 79-80 motif is commonly reversed as Phe79-Tyr80 in insect DHPAAS, and these residues also surround the external aldimine of the PLP-substrate complex (Fig. 3). Therefore we hypothesized that these aromatic residues may also be involved in the DHPAAS catalytic mechanism and should be useful for the classification of DHPAAS. Within the five identified DHPAAS groups, Phe79-Tyr80 is conserved in Apis and mosquito. In Drosophila DHPAAS sequences annotated as isoform X1, Phe79-Tyr80 is conserved, whereas in those annotated as isoform X2 (includes NP_476592.1 27 ), Tyr79-Tyr80 is conserved. Lepidoptera and formicidae DHPAAS groups contain a mixture of Phe79-Tyr80, Tyr79-Tyr80 and Tyr79-Phe80.
In this study, the B. mori sequence XM_004930959.2 was selected as a distinct DHPAAS sequence containing all three identified DHPAAS specific residues Phe79, Tyr80 and Asn192, as well as Gly353, which is reported to increase substrate specificity to L-DOPA 26 . In addition, Phe79Tyr, Tyr80Phe and Asn192His DHPAAS catalytic variants were designed to explore tuning the bifunctional production of dopamine and DHPAA.
Switching the mechanism of DHPAAS. The native sequence of wild-type B. mori XM_004930959.2 was synthesized and cloned into the E. coli expression vector pE-SUMO. DHPAAS was purified by metal affinity chromatography according to the methods section. DHPAAS activity was screened by separation of substrates and products using thin-layer chromatography (TLC). Enzymatic products were verified by LC-MS operated in multiple reaction monitoring (MRM) mode.
The recombinant B. mori XM_004930959.2 protein produced DHPAA as the major product of L-DOPA, as indicated by the identification of negative ion m/z 151.10 and lack of major dopamine ions. The identification B. mori DHPAAS indicates that the above analysis of DHPAAS phylogenetic groups is accurate. Structural analysis supports a hypothesis that the Phe79Tyr-Tyr80Phe-Asn192His triple variant would have DDClike activity, while Asn192His and Phe79Tyr-Ty80Phe variants would result in a mixture of DHPAAS and DDC activities. To test this hypothesis, and gain insight into a more comprehensive DHPAAS mechanism, the enzymatic activities of Phe79Tyr, Tyr80Phe, and Asn192His DHPAAS variants were evaluated (Figs 5 and 6).
Dopamine, which can be easily observed using TLC with ninhydrin staining, was identified as the major product of the Phe79Tyr-Tyr80Phe-Asn192His DHPAAS variant, supporting the above hypothesis (Supplementary Figure 4). After longer incubation times, THP was detected as the highest intensity positive ion of Phe79Tyr-Tyr80Phe-Asn192His DHPAAS reaction with L-DOPA (Fig. 5b). Comparison of DHPAAS activities via H 2 O 2 product quantification indicates that the Asn192 residue is most important for maintaining DHPAAS activity, while Phe79 and Tyr80 residues also contribute to the DHPAAS mechanism (Fig. 6, Table 1). Characterization of aromatic products from all four DHPAAS variants was then performed in vitro.
Direct production of THP via engineered DHPAAS. After confirming that THP could be directly produced by Phe79Tyr-Tyr80Phe-Asn192His DHPAAS, in vitro production was quantified using all four designed B. mori DHPAAS variants. Dopamine, DHPAA and THP were monitored using LC-MS operated in MRM mode (Fig. 7a-c). THP was highly sensitive to oxidation, as indicated by the detection of ion m/z 284.10, which corresponds to THP-quinone ([THP-3H] + = 284.0917).
In vitro THP production improved significantly after supplementation with ascorbate to overcome product degradation via H 2 O 2 mediated oxidation. After supplementation with 2.5 mM sodium ascorbate, conversion of L-DOPA to THP increased to 23% (219 µM) using Phe79Tyr-Tyr80Phe-Asn192His DHPAAS. This surpasses the highest bioconversion of dopamine to THP reported at 15.9% 10 . Ascorbic acid did not inhibit DHPAA production by DHPAAS indicating that DHPAA is a direct enzymatic product of L-DOPA, rather than a secondary product of dopamine oxidation by H 2 O 2 .
As predicted, overall DHPAA production was highest with wild-type DHPAAS and the Phe79Tyr-Tyr80Phe variant, and lowest with the Asn192His and Phe79Tyr-Tyr80Phe-Asn192His variants (Fig. 7a-c, Table 1). The expected opposite trend was observed for dopamine production, which was highest with the Phe79Tyr-Tyr80Phe-Asn192His variant and lowest with wildtype DHPAAS, while dopamine production by the Asn192His variant was higher than that of the Phe79Tyr-Tyr80Phe variant. Therefore, the in vitro results support the above structure-based hypothesis about the effect of Phe79, Tyr80 and Asn192 on DHPAAS mechanism switching (Table 1, Fig. 8). Additional testing was performed to further optimize DHPAAS-mediated THP bioproduction in vivo.
Tuning DHPAA-mediated BIA bioproduction. Initial attempts at bioproduction using E. coli grown in lysogeny broth (LB) medium resulted in low THP titers, with higher amounts produced by the Phe79Tyr-Tyr80Phe variant, followed by  that of wild-type DHPAAS. After changing the growth medium from LB to M9 minimal medium, THP production increased ( Fig. 7d-f).
In contrast to the in vitro results, the Phe79Tyr-Tyr80Phe variant was the best relative in vivo THP producer of this study, with titers exceeding 1 µM (0.287 mg/L). Wild-type DHPAAS produced the next highest amounts of in vivo THP, followed by Phe79Tyr-Tyr80Phe-Asn192His DHPAAS as the third-best THP producer. The Asn192His variant produced the lowest in vivo THP titers when including ascorbate. A diastereomeric mixure of (R,S)-THP was produced by DHPAAS in vivo as demonstrated by chiral LC-MS analysis (Supplementary Figure 6). After addition of Coptis japonica 6-OMT, CNMT and 4-OMT genes (Fig. 1) to DHPAAS expressing E. coli, THP and reticuline production followed similar trends. The Asn192His and Phe79Tyr-Tyr80Phe-Asn192His DHPAAS variants resulted in higher reticuline titers than THP titers suggesting high conversion of THP to reticuline. However, reticuline titers were limited to approximately 0.2 µM using all four DHPAAS variants, which indicates that there is a bottleneck in the conversion of (R,S)-THP to reticuline under the conditions tested (Fig. 7g, h).
To improve the limited reticuline titers, additional DHPAAS, DDC, 6-OMT, and CNMT genes were tested. Addition of a bacterial DDC resulted in an increase in THP titers to above 2 µM (0.57 mg/L), accompanied by a huge increase in dopamine levels. However, reticuline titers did not simply increase dependent on THP concentration, with the bottleneck appearing strongest at the 6-OMT and 4-OMT steps. Expression of DHPAAS-Phe79Tyr-Tyr80Phe-Asn192His together with wild-type  Km, kcat, and Vmax values were calculated with Prism 7 using H2O2 production data from Fig. 6. 95% confidence intervals listed in parenthesis (n ≥ 8 independent measurements). DDC activity Vmax is calculated based on the first in vitro measurements of dopamine product (Fig. 7a, t = 9 min.). Estimated % DHPAAS activity is based on the comparison of these two rates.
DHPAAS, as well as Coptis japonica 6-OMT, CNMT, and 4-OMT, resulted in a moderate increase in relative THP. Yet, the bottleneck to reticuline still remained, although it appeared to be relaxed at the 6-OMT and 4-OMT steps. Metabolomics analysis suggested that conversion of (R,S)-THP to reticuline may not be optimal during simultaneous conversion of L-DOPA to THP due to disruption of pathways relating to SAM cofactor recycling (Fig. 9a). Therefore a two-step cell production system was tested as described in the methods section 10,11 , in an attempt to balance the two potentially counteracting pathways. This resulted in a four-to eight-fold increase of (R,S)-THP titer to 9.45 µM (2.71 mg/L) in the first step via wild-type DHPAAS and DHPAAS-Phe79Tyr-Tyr80Phe-Asn192His. A second addition of BL21(DE3) containing 4-OMT from C. japonica together with 6-OMT and CNMT from P. somniferum was then able to relieve the bottleneck to produce 3.7 μM 3-hydroxycoclaurine (3HC), 1.4 μM 3-hydroxy-N-methylcoclaurine (3HNMC) and 1.5 μM reticuline after an additional 18.5 h (Fig. 9b, c). This represents a 7.5-fold increase in reticuline, as well as additional increases in 3HC and 3HNMC, demonstrating scalable titers using the DHPAASmediated system.

Discussion
To scale up THP and reticuline production, improved synthetic biology design can be driven by two types of learning information: enzyme parameters and metabolomics data. Regarding learning for enzyme improvement, Phe79, Tyr80, and Asn192 were all elucidated as key residues that can tune the DHPAAS mechanism for optimized THP precursor ratios. Yet questions still remain about the exact function of these residues, including their role in O 2 activation, which remains a somewhat mysterious activity of PLP enzymes 28  DHPAAS specificity for L-DOPA was high, with no observed activity for tyrosine or phenylalanine by the Phe79Tyr-Tyr80Phe-Asn192His variant. This indicates that competitive inhibition by aromatic amino acids is not a major consideration for DHPAAS-mediated biosynthesis. However, preliminary analysis indicates possible feedback inhibition of DHPAAS by dopamine and DHPAA, and possible substrate inhibition at high concentrations of L-DOPA. These factors, together with the low measured Km of B. mori DHPAAS, may help to explain why L-DOPA was not consumed during in vivo bioproduction experiments. It is also possible that DHPAAS is inhibited by unknown growth medium components. Therefore, substrate concentration, substrate specificity and inhibition should be comprehensively characterized as data for enzyme learning.
During in vitro production of THP, high concentration dopamine was best, as more dopamine can quickly consume reactive DHPAA while it is being produced. Thus, Asn192His and Phe79Tyr-Tyr80Phe-Asn192His variants were observed to produce the highest THP levels in vitro. However, in vivo environments include many more cellular and chemical components that can compete with dopamine for reaction with DHPAA. Accordingly, the use of minimal medium together with higher levels of DHPAA produced by wild-type DHPAAS and the Phe79Tyr-Tyr80Phe variant resulted in better in vivo THP titers. These results highlight that the optimal dopamine to DHPAA ratio for THP production depends on the culture conditions.
The best reported THP titer of 1 mM (287 mg/L) was produced from 12.6 mM dopamine using MAO expressing cells 10 . MAOmediated conversion of dopamine to THP was 15.9%, without reported metrics for conversion of L-DOPA to dopamine 10 . In the current study, DHPAAS expressing cells produced 9.45 μM THP (2.71 mg/L) with a yield of 1.89% from 1 mM L-DOPA. The lower titer and yield may be explained by the 12.6-fold lower substrate concentration and the additional decarboxylation step from L-DOPA. Moreover, a majority of 1 mM L-DOPA remained unconsumed in the culture medium, suggesting low substrate permeability and possibly higher intracellular yields. L-DOPA utilization by DHPAAS is much better in vitro, with 23% conversion of L-DOPA to 219 µM THP. In the MAO-mediated system, a THP titer of 99.2 µM resulted in the highest reticuline titers 11 . Therefore, in vitro results suggest that DHPAAS-mediated THP production is capable of reaching optimal levels for downstream BIA production. However, DHPAAS-mediated in vivo production needs to be improved to reach the 1 mM THP benchmark and computational yields greater than 50%.
The DHPAAS-mediated pathway also requires improvements in downstream BIA production. After the establishment of proofof-principle bioproducers, it usually requires more than five years of optimization to reach industrial demands 30 Fig. 8 General scheme of tuning engineered DHPAAS variants for THP production. Based on the activities presented in Table 1, major products are presented as larger sized molecules relative to the scale of minor products ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-09610-2 complete pathway is now assembled into a single E. coli cell to produce high titer reticuline from glucose 31 , 10 years after the development of the first MAO-mediated reticuline producing system 32 . To better match this benchmark, a complete DHPAASmediated pathway from glucose to (S)-reticuline is now under development. Metabolomics-directed learning data should include flux from toxic intermediates DHPAA 20 and THP 10 towards less-toxic reticuline. Addition of NCS can increase conversion of DHPAA to selectively produce (S)-THP, leading to better substrate conversion by SAM dependent BIA methyltranferases 31,33 . A related OMT bottleneck with impaired SAM cofactor recycling was reported for vanillin biosynthesis in E. coli grown on M9 minimal medium 34 . This supports balancing of SAM recycling pathways as another strategy to improve reticuline production (Fig. 9). Quenching of H 2 O 2 by antioxidant enzymes including catalase II (katE) should also be pursued to increase BIA titers.
Additional M-path analysis was performed to search alternative bioproduction pathways that might be improved using DHPAAS or AAS, linking the learning process of enzyme database improvement to new cycles of pathway design. For example, M-path identified EC 4.2.1.107-109 as enzymes to mediate biosynthesis of 2'-norberbamunine 35 , homovanillic acid, imidazole-4-acetaldehyde, and 4-amino-phenylacetaldehyde. In conclusion, this study demonstrates that functional enzyme engineering can be effectively applied to high value bioproduction pathways. Most importantly, further development of the DHPAAS-mediated pathway is required to improve downstream BIA production in future studies. Computational prediction of enzyme-mediated THP production. Mathematical models were constructed for the nonsymmetrical (DDC-MAO) and symmetrical (DDC-DHPAAS) pathways. For the nonsymmetrical pathway, since MAO recognizes a variety of amines, competitive inhibition was introduced into the MAO reaction velocity V MAO 12,13 . Models were constructed with and without product feedback inhibition factored into the reaction velocities of MAO (V MAO ), DDC (V DDC ) and DHPAAS (V DHPAAS ). Drain of DHPAA was later included in some models. Possible ranges of parameter values were surveyed and collected from literature and BRENDA 15 . In order to predict the performance of each pathway, parameter values were randomly generated within the set ranges, and Monte-Carlo simulation was carried out according to Supplementary Methods. The iteration number was 10,000, and simulation time was 0-50 h. L-DOPA was fed as a constant term based on randomly generated parameters. When the maximum amount of 100 mM L-DOPA was reached, substrate feeding to the system was stopped. A homemade program was implemented in Python 3.0 using scipy.integrate.odeint as the solver for numerical simulations. More detailed computational methods are described in Supplementary Methods, Supplementary  Figures 2-3, and Supplementary Table 2.
Homology modeling and phylogenetic analysis. Dimeric homology models of B. mori DHPAAS and P. somniferum TyDC1 were generated with MODELLER 22 operated within Chimera 23 . Crystal structures of D. melanogaster DDC (PDB ID: 3K40) 25 and histidine decarboxylase (4E1O) 26 were used as templates for B. mori DHPAAS modeling. The structure of Sus Scrofa DDC in complex with carbidopa (PDB ID: 1JS3) 24 was used as a template for TyDC1. Binding of PLP covalently linked to aromatic amino acid substrate, and structure refinement were performed with Molecular Operating Environment (MOE). Completed structures were analyzed in PyMOL.
For phylogenetic analysis, insect AAAD and DHPAAS sequences were collected from the protein BLAST non-redundant database after searching from insect sequences NP_476592.1, NP_724162.1, XP_319838.3, EDS39158.1, EAT37246.1, and EAT37247.1. Duplicate sequences, sequences containing X as an amino acid, and sequences over 700 amino acids in length were all removed. The resulting sequences were aligned and a phylogenetic tree was generated using a split value of 0.12. After removing partial sequences, clusters were identified from a sequence similarity table generated by MOE.
Preparation of recombinant B. mori DHPAAS. The full length B. mori DHPAAS native sequence was synthesized by GeneArt (Invitrogen) and cloned into the pE-SUMO vector with kanamycin resistance (LifeSensors Inc.) via BsaI restriction sites. Mutations were generated using overlapping PCR with primers shown in Supplementary Table 4. DHPAAS expression vectors were transformed into BL21 (DE3) maintained in LB supplemented with 50 µg/mL kanamycin, or BL21(DE3) pLysS maintained in LB supplemented with 50 µg/mL kanamycin and 34 µg/mL chloramphenicol. Expression of recombinant DHPAAS was induced by the addition of 0.2-0.45 mM IPTG to E. coli grown aerobically in LB medium. After induction, culture temperature was lowered to 14-16°C. After overnight incubation, cells were pelleted by centrifugation, resuspended in phosphatebuffered saline (PBS) and lysed by sonication while being cooled over ice. After centrifugation, clarified lysates were loaded onto HiTrap TALON or HisTrap HP Step 1 23 h Step 2  Metabolomics directed optimization of (R,S)-THP and reticuline production. a Monitored metabolites including homocysteine (HCys) and glutathione (GSH) are listed in black text, whereas unmeasured metabolites are listed in graey. Lower relative levels of 3-hydroxycoclaurine (3HC) and reticuline suggested that a bottleneck was strongest at 6-OMT and 4-OMT steps, which are inhibited by S-adenosylhomocysteine (SAH) as illustrated by the red flat-headed curves 33 . Lower utilization of S-adenosylmethionine (SAM) and higher accumulation of SAH accompanied stronger THP to reticuline bottlenecks. Desired pathways mediated by DHPAAS, 6-OMT, CNMT, 4-OMT, methionine synthase (MS), and methionine adenosyltransferase (MAT) are shown in blue. b, c Titers of THP, 3HC, 3hydroxy-N-methylcoclaurine (3HNMC), and reticuline from 1 mM L-DOPA using a two-step cell process as described in the methods section. THP was quantified 23 h after L-DOPA adddition and three downstream BIAs were quantified 18.5 h after addition of the second bioproducer. Here, error bars define the standard error of the mean (n = 3 independent measurements). Source data of Fig. 9 panels b and c are provided in a Source Data file NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-09610-2 ARTICLE columns (GE Life Sciences). After loading, columns were washed with PBS and 10-20 mM imidazole. Recombinant DHPAAS was eluted with 450-1000 mM imidazole. Buffer was changed to PBS supplemented with PLP using Amicon Ultra-15 centrifugal filters (Millipore).
TLC analysis of DHPAAS reactions. TLC was performed on aluminum plates coated with silica gel 60 F254 (Merck Millipore). A mixture of 1-butanol, acetic acid and H 2 O at a ratio of 7:2:1 was used as the mobile phase. DHPAAS reaction components were analyzed under UV light before ninhydrin staining.
Mass analysis of DHPAAS substrates and products. Substrate and products were identified with Shimadzu LCMS-8050 and LCMS-8060 ESI triple quadrupole mass spectrometers. Quantitative analysis was performed using the Shimadzu LCMS systems operated in multiple reaction monitoring (MRM) mode with the following parameters: ionization, ESI; DL temp., 250°C; block heater temp., 400°C; interface temp., 300°C; nebulizing gas flow, 3.0 L/min; drying gas flow, 10.0 L/min; heating gas flow, 10.0 L/min. The ion source was connected to a Shimadzu Nexera X2 UHPLC system 36  H 2 O 2 production was analyzed in 96 well plates using a fluorometric hydrogen peroxide assay kit (Sigma). 0.6-0.8 µg soluble DHPAAS in PBS (20 µL) was mixed with varying concentrations of L-DOPA (10 µL), followed by the addition of 30 µL peroxidase enzyme and fluorescent substrate mix (Sigma). Fluorescence was detected with a SpectraMax Paradigm microplate reader (Molecular Devices). Kinetic data included eight independent conditions for wild-type DHPAAS (one outlier), Asn192His variant (two outliers), and Phe79Tyr-Tyr80Phe variant, while 15 independent conditions were used for the Phe79Tyr-Tyr80Phe-Asn192His variant. Kinetic data was analyzed using the kcat function of Prism 7 with outlier elimination.
In vivo DHPAAS-mediated THP bioproduction. DHPAAS sequences were PCR amplified with primers containing NcoI and XhoI restriction sites (Supplementary Table 4) for cloning into pTrcHis2B. The resulting untagged expression vectors were then transformed into BL21(DE3)pLysS. For bioproduction, cells were grown in M9 medium supplemented with 2% glucose, 15.6 mM sodium ascorbate, 100 µg/mL ampicillin, and 34 µg/mL chloramphenicol. Cultures containing 3.5 mL medium were grown at 37°C with shaking. When culture OD 600 reached 0.2-0.4, DHPAAS expression was induced with 0.97 mM IPTG, and the growth temperature was lowered to 25°C. At 1 h, 13 min post-induction, 3.4 mg L-DOPA (0.97 mg/ mL, 4.92 mM) was then added to each culture, followed by addition of PLP to a final concentration of 4.86 µM. Culture temperature was decreased to 16°C 12.9 h. after the addition of L-DOPA. Culture samples (300-500 µL) for four time points were filtered through Amicon Ultra 0.5 mL centrifugal filters (Millipore) with a molecular weight cut-off of 3000 Da. Approximately 4-5 mg ascorbate was added to each culture 22.7 h. after substrate addition, and cultures were moved to 4°C. Finally, 49.8 h. after substrate addition, cultures were centrifuged at 4500 × g, followed by collection and analysis of supernatants. Culture samples were diluted in methanol with camphor sulfonic acid and ascorbate for quantification of dopamine, DHPAA and THP. Individual dilutions were analyzed three times for each condition, and in vivo figures with error bars were generated using Prism 7.
In vivo DHPAAS-mediated reticuline bioproduction. pTrcHis2B containing DHPAAS was transformed, together with a pACYC184 derived vector containing C. japonica 4-OMT, CNMT, and 6-OMT, into BL21(DE3). Resulting E. coli reticuline bioproducers were selected with 100 µg/mL ampicillin and 34 µg/ mL chloramphenicol. Reticuline production was tested in M9 minimal medium supplemented with 2% glucose. Cells were grown to an OD 600 of 0.2-0.3 before the addition of 0.5 mM IPTG, 450 µM L-DOPA, and 4.54 mM sodium ascorbate. 17.2 h. after substrate addition PLP was added to 54.8 µM along with a second addition of an additional 444 µM ascorbate. SAM was added 23.5 h after substrate addition to a final concentration of 9.7 µM. Reticuline was produced at 25°C with shaking. Filtered culture medium was diluted in acetonitrile with camphor sulfonic acid and ascorbate for quantification of dopamine, DHPAA, THP, and reticuline. Single cultures of Phe79Tyr-Tyr80Phe and Phe79Tyr-Tyr80Phe-Asn192His DHPAAS containing E. coli were tested. Duplicate cultures of wild-type and Asn192His DHPAAS containing E. coli were tested. Each dilution was measured in duplicate, resulting in duplicate measurements for Phe79Tyr-Tyr80Phe and Phe79Tyr-Tyr80Phe-Asn192His DHPAAS-mediated reticuline production, and quadruplicate measurements for wild-type and Asn192His DHPAAS-mediated reticuline production. In vivo figures with error bars were generated using Prism 7.
Two-step cell production of BIA. BL21(DE3) was transformed with wild-type DHPAAS in pTrcHis2B, DHPAAS-Phe79Tyr-Tyr80Phe-Asn192His in pE-SUMO, and C. japonica 4-OMT, CNMT and 6-OMT in pACYC184. For the first step of THP production, this three plasmid system was initially grown at 37°C in TB with no glycerol supplemented with 1.5% glucose, 100 µg/mL ampicillin, and 50 µg/mL kanamycin. After reaching OD 600 0.38, IPTG was added to a final concentration of 0.5 mM IPTG. 1.5 h after induction, the temperature was lowered to 25°C. At 5.5 h post-induction, cells were collected by centrifugation at 4000 g, stored overnight at −80°C, and the pellet from~43 mL culture was resuspended in M9 with reduced calcium, 0.2% Triton X-100, 1.5% glucose, 10 μM PLP, 10 mM sodium ascorbate, and 1 mM L-DOPA, to a final volume of 6.5 mL. After mixing, cultures were kept at 24-25°C for 1.5 h and then centrifuged at 5000 × g to assist condensation of DHPAA with dopamine in the supernatant. Twenty-five hours after substrate addition, the static culture was centrifuged again at 5000 × g and the THP containing supernatant was used for the next downstream BIA production step.
For the second step of increased downstream BIA production, pET23a containing 4-OMT from C. japonica together with 6-OMT and CNMT from P. somniferum was transformed into BL21(DE3). This second BIA producer was also initially grown at 37°C in TB with no glycerol supplemented with 1.5% Glucose and 100 µg/mL ampicillin. After reaching OD 600 0.78, IPTG was added to a final concentration of 0.5 mM. 1.5 h after induction, the temperature was lowered to 25°C. 5.5 h post-induction, cells were collected by centrifugation at 4000 × g, stored for 2 nights at −80°C, and the pellet from~46 mL culture was resuspended in the supernatant from the first step. Downstream BIA production was then tested at 25°C with shaking. Diluted samples of culture medium were analyzed with three independent measurements for each condition.
Reporting summary. Further information on experimental design is available in the Nature Research Reporting Summary linked to this article.

Data availability
Data supporting the findings of this work are available within the paper and its Supplementary Information files. A reporting summary for this Article is available as a Supplementary Information file. The datasets generated and analyzed during the current study are available from the corresponding author upon reasonable request. The source data underlying Figs. 6, 7, 9b, and 9c, as well as Supplementary Figure 5 are provided as a Source Data file.