Microbial Production of Methyl Anthranilate, a Grape Flavor Compound

Microbial production of methyl anthranilate, a grape flavor compound

Zi Wei Luoa,b,1, Jae Sung Choa,b,1, and Sang Yup Leea,b,c,d,2

aMetabolic and Biomolecular Engineering National Research Laboratory, Department of Chemical and Biomolecular Engineering (BK21 Plus Program), Institute for the BioCentury, Korea Advanced Institute of Science and Technology, 34141 Daejeon, Republic of Korea; bSystems Metabolic Engineering and Systems Healthcare Cross-Generation Collaborative Laboratory, Korea Advanced Institute of Science and Technology, 34141 Daejeon, Republic of Korea; cBioProcess Engineering Research Center, Korea Advanced Institute of Science and Technology, 34141 Daejeon, Republic of Korea; and dBioInformatics Research Center, Korea Advanced Institute of Science and Technology, 34141 Daejeon, Republic of Korea

Contributed by Sang Yup Lee, April 5, 2019 (sent for review March 6, 2019; reviewed by Jay D. Keasling and Blaine A. Pfeifer) Methyl anthranilate (MANT) is a widely used compound to give While these biotransformation procedures are considered more grape scent and flavor, but is currently produced by petroleum-based natural and ecofriendly compared with chemical synthesis, their processes. Here, we report the direct fermentative production of actual use is limited due to low yields, long reaction times, and MANT from glucose by metabolically engineered Escherichia coli and formation of byproducts (5). In addition, the chemical and bio- Corynebacterium glutamicum strains harboring a synthetic plant- transformation processes mentioned above depend on substrates of derived metabolic pathway. Optimizing the key enzyme anthranilic petroleum origin. For these reasons, we were motivated to produce acid (ANT) methyltransferase1 (AAMT1) expression, increasing the MANT through one-step microbial fermentation of renewable direct precursor ANT supply, and enhancing the intracellular avail- abilityandsalvageofthecofactorS-adenosyl-L-methionine required feedstocks (e.g., glucose), which would offer 100% biobased nat- by AAMT1, results in improved MANT production in both engineered ural MANT in an ecofriendly manner. microorganisms. Furthermore, in situ two-phase extractive fermen- To our knowledge, there have been rare attempts on the de tation using tributyrin as an extractant is developed to overcome novo microbial production of MANT, except for two reports MANT toxicity. Fed-batch cultures of the final engineered E. coli nearly 30 y ago describing MANT biosynthesis from simple sugars and C. glutamicum strains in two-phase cultivation mode led to (i.e., maltose) by the wild-type fungi, Poria cocos (6) and Pycno- SCIENCES the production of 4.47 and 5.74 g/L MANT, respectively, in minimal

porus cinnabarinus (7). Unfortunately, the productivities achieved APPLIED BIOLOGICAL media containing glucose. The metabolic engineering strategies de- in these two studies were extremely low (18.7 mg/L MANT pro- veloped here will be useful for the production of volatile aromatic duced after 5 d of culture). Also, the underlying biosynthetic esters including MANT. mechanisms, including the biosynthesis genes, enzymes, and pathways, in these two fungal species have not been elucidated. metabolic engineering | Escherichia coli | Corynebacterium glutamicum | Thus, it is a prerequisite to first identify a metabolic pathway methyl anthranilate | two-phase fermentation leading to the biosynthesis of MANT from simple carbon sources (e.g., glucose), before implementing various metabolic engineering ethyl anthranilate (MANT), which gives grape scent and Mflavor, has been extensively used in flavoring foods (e.g., strategies to develop microbial strains capable of efficiently pro- candy, chewing gum, soft drinks, and alcoholic drinks, etc.) and ducing MANT based on the reconstituted biosynthetic pathway. drugs (as a flavor enhancer and/or mask). Due to its pleasant aroma, MANT is an important component in perfumes and cos- Significance metics. MANT also has other important industrial applications as a bird and goose repellent for crop protection, as an oxidation in- Methyl anthranilate (MANT) is widely used in the flavoring and hibitor or a sunscreen agent, and as an intermediate for the syn- cosmetics industry to give grape scent and flavor. In an effort thesis of a wide range of chemicals, dyes, and pharmaceuticals (1). to replace the conventional petroleum-based synthesis of MANT is a natural metabolite giving the characteristic odor of MANT, we report the direct fermentative production of MANT Concord grapes and occurs also in several essential oils (e.g., from glucose in metabolically engineered Escherichia coli and neroli, ylang ylang, and jasmine) (1). It has been challenging and Corynebacterium glutamicum strains. A synthetic plant-derived economically infeasible to directly extract MANT from these metabolic pathway was introduced and extensive metabolic plants due to low yields. Currently, MANT is commercially engineering was performed including fine-tuning key enzyme manufactured by petroleum-based chemical processes, which levels and increasing the availability of precursor and cofactor mainly rely on esterification of anthranilic acid (ANT) with metabolites. A two-phase extractive cultivation was developed methanol or isatoic anhydride with methanol, using homogeneous using an extractant solvent to recover MANT in situ, which led acids as catalysts (2). These processes, however, suffer from sev- to high levels of MANT production. This work demonstrates a eral disadvantages, for example, the requirement of acid catalysts promising sustainable alternative to MANT production and in large quantities and problems with disposal of these toxic liquid presents strategies applicable toward production of other acids after the reaction (2). Moreover, MANT produced by such valuable natural compounds. chemical methods is labeled “artificial flavor,” which does not meet the increasing demand by consumers for natural flavors. Author contributions: S.Y.L. designed research; Z.W.L. and J.S.C. performed research; Taking another important flavoring agent vanillin as an example, Z.W.L., J.S.C., and S.Y.L. analyzed data; and Z.W.L., J.S.C., and S.Y.L. wrote the paper. market preference for natural vanillin has led to a far higher price Reviewers: J.D.K., University of California, Berkeley; and B.A.P., University at Buffalo. of $1,200–$4,000/kg over $15/kg for artificial vanillin (3). Such a The authors declare no conflict of interest. market for natural MANT is also highly desirable, but unfortu- Published under the PNAS license. nately there have so far been no promising methods for preparing 1Z.W.L. and J.S.C. contributed equally to this work. MANT from natural sources and/or by natural means. Several 2To whom correspondence should be addressed. Email: [email protected]. enzymatic and microbial whole-cell biotransformation approaches This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. have been attempted for MANT production by esterification of 1073/pnas.1903875116/-/DCSupplemental. ANT (4) or N-demethylation of N-methyl methyl anthranilate (5).

www.pnas.org/cgi/doi/10.1073/pnas.1903875116 PNAS Latest Articles | 1of8 Downloaded by guest on October 2, 2021 In this study, we report the development of metabolically synthetic metabolic pathway originated from plant for de novo engineered Escherichia coli and Corynebacterium glutamicum MANT biosynthesis was constructed in both E. coli and C. gluta- strains capable of producing MANT directly from glucose micum. Next, MANT production in both engineered organisms through fermentation (Fig. 1). E. coli was initially chosen as a was improved through optimization of the key enzyme level, in- model organism for metabolic engineering toward efficient crease of flux to the direct precursor metabolite, increase of the production of MANT. In addition, C. glutamicum, a generally availability of cosubstrate required in MANT synthesis, and es- recognized as safe (GRAS) strain, was also engineered for the tablishment of in situ two-phase extractive cultivation process. production of MANT for its consequent human consumption to Finally, two-phase fed-batch cultures of the best E. coli and C. give grape flavor and scent in food and cosmetics industries. We glutamicum strains were performed to demonstrate their po- applied multiple strategies to produce MANT by optimizing its tential for large-scale production of MANT from glucose in biosynthesis in both E. coli and C. glutamicum (Fig. 2). First, a minimal media.

Fig. 1. The metabolic network related to MANT biosynthesis from glucose in (A) E. coli and (B) C. glutamicum, as well as metabolic engineering strategies employed in this study. Abbreviations: ANT, anthranilate; ASP, L-aspartate; CHA, chorismate; DAHP, 3-deoxy-D-arabinoheptulosonate 7-phosphate; DHQ, 3- dehydroquinate; DHS, 3-dehydroshikimate; E4P, erythrose 4-phosphate; EPSP, 5-enolpyruvyl-shikimate 3-phosphate; G6P, glucose 6-phosphate; Gln, glutamine; Glu, glutamate; HCYS, L-homocysteine; ILE, isoleucine; L-PHE, L-phenylalanine; L-TRP, L-tryptophan; L-TYR, L-tyrosine; MANT, methyl anthranilate; MET, L-methionine; PCA, protocatechuate; PEP, phosphoenolpyruvate; PPP, pentose phosphate pathway; PRANT, N-(5-phosphoribosyl)-anthranilate; PTS, phosphotransferase system; PYR, pyruvate; QA, quinate; S3P, shikimate-3-phosphate; SAH, S-adenosyl-L-homocysteine; SAM, S-adenosyl-L-methionine; SER, L-serine; SHK, shikimate; SRH, S-ribosyl-L-homocysteine; TCA, tricarboxylic acid cycle. Genes that encode enzymes: ppsA, PEP synthetase; pykF, PYR kinase I; pykA,PYRkinaseII;tktA,transketolase I; aroGfbr, feedback-inhibition resistant mutant of DAHP synthase; aroB, DHQ synthase; aroD, DHQ dehydratase; aroE and ydiB, SHK dehydrogenase; aroK, SHK kinase I; aroL, SHK kinase II; aroA, 3-phosphoshikimate-1-carboxyvinyltransferase; aroC, CHA synthase; tyrA and pheA, TyrA and PheA subunits of CHA mutase, respectively; trpEfbr and trpD, ANT synthase component I (feedback inhibition-resistant mutant) and II, respectively; glnA, Gln synthetase; cysEfbr,SER acetyltransferase (feedback inhibition-resistant mutant); metAfbr, homoserine O-succinyltransferase (feedback inhibition-resistant mutant); luxS, S-ribosylhomocysteine lyase; mtn,5′-methylthioadenosine/S-adenosylhomocysteine nucleosidase; metK, MET adenosyltransferase; aamt1, ANT methyltransferase1; pgi,G6Pisomerase; zwf, G6P 1-dehydrogenase; tkt, transketolase; hdpA, dihydroxyacetone phosphate phosphatase; qsuB, DHS dehydratase; qsuD, QA/SHK dehydrogenase; metB, cystathionine-γ-synthase; sahH,SAHhydrolase;mcbR and Ncgl2640, transcriptional regulator. The solid arrows indicate single metabolic reaction, and the dashed arrows indicate multiple reactions. Overexpressed genes are marked in red, and red X indicates gene deletions. In situ two-phase extractive cultivation process is schematically illustrated, in which the MANT produced in the aqueous phase is extracted into the organic phase.

2of8 | www.pnas.org/cgi/doi/10.1073/pnas.1903875116 Luo et al. Downloaded by guest on October 2, 2021 SCIENCES APPLIED BIOLOGICAL

Fig. 2. The overview of engineering strategies to optimize MANT production in E. coli and C. glutamicum.(A) Engineering strategies applied to produce MANT from glucose in E. coli, which include (1) initial construction of the de novo MANT biosynthesis pathway, (2) establishing an in situ two-phase extractive cultivation process to address MANT toxicity, (3) optimizing the key enzyme AAMT1 expression, (4) increasing the precursor ANT supply by metabolic engineering, (5) enhancing the availability of cosubstrate SAM required by the AAMT1-catalyzed reaction, and (6) in situ two-phase extractive fed-batch culture for bioreactor-scale MANT production. The successive increase in MANT titer with each strategy is also illustrated. The pink bars represent the concentration of MANT (in grams per liter). Values and error bars represent means and SDs of biological duplicates. The white circles represent individual data points. Error bars represent mean ± SD (n = 2). The P values were computed by two-tailed Student’s t test (**P < 0.01; ***P < 0.001; ****P < 0.0001). (B) Engineering strategies to produce MANT from glucose using C. glutamicum, which include (1) initial construction of the de novo MANT biosynthesis pathway, (2) optimizing the key enzyme AAMT1 expression, (3) increasing precursor ANT supply by metabolic engineering, (4) establishing an in situ two-phase extractive cultivation process to address MANT toxicity, (5) salvaging the availability of cosubstrate SAM required by the AAMT1-catalyzed reaction, and (6) in situ two- phase extractive fed-batch culture for bioreactor-scale MANT production. The successive increase in MANT titer with each strategy is also illustrated. The pink bars represent the concentration of MANT (in grams per liter). Values and error bars represent means and SDs of biological triplicates. The white circles represent individual data points. Error bars represent mean ± SD (n = 3). The P values were computed by two-tailed Student’s t test (****P < 0.0001).

Luo et al. PNAS Latest Articles | 3of8 Downloaded by guest on October 2, 2021 Results and Discussion Using this two-phase system in shake-flask cultivation (a Constructing a MANT Biosynthetic Pathway in E. coli. Due to the lack 5:1 aqueous-to-organic phase ratio) (SI Appendix,Fig.S4A), of knowledge on the MANT metabolism in the two fungal species W3110 trpD9923 strain harboring pTrcT produced 65.6 ± 0.4 mg/ described above, we focused on the mechanism of MANT bio- LMANT(Fig.2A and SI Appendix, Fig. S4B;seeSI Appendix, synthesis in plants to identify a MANT biosynthetic pathway. In- Supplementary Information Materials and Methods for the cal- triguingly, literature survey (8) suggested two different MANT culation of MANT concentration), which was 83.2% higher than biosynthetic mechanisms present in plants (SI Appendix,Fig.S1). that obtained in the single-phase culture. Also, almost all of the Both routes share the same precursor metabolite ANT, which produced MANT was extracted into the solvent phase, while is derived from the L-tryptophan (L-TRP) biosynthesis pathway MANT was undetectable in the aqueous phase (SI Appendix, in plants. For further conversion of ANT to MANT, one route Fig. S4B). involves two reaction steps: CoA activation catalyzed by anthranilate-CoA ligase followed by acyl transfer by anthraniloyl- Optimizing AAMT1 Expression for MANT Production in E. coli. Fol- CoA:methanol acyltransferase using CoA, ATP, and methanol as lowing the establishment of in situ two-phase extractive cultivation cosubstrates. The other route uses a single-step conversion of process, we set out to optimize MANT production by manipulating L expression of the key enzyme AAMT1 (Fig. 1A). Expression levels ANT to MANT catalyzed by S-adenosyl- -methionine (SAM)- opt dependent methyltransferase. In this study, we chose the latter of AAMT1 were varied by expressing the aamt1 gene, either on route as it is a simpler reaction requiring less cofactors, and more the medium-copy pTrc99A vector under a set of six synthetic consti- importantly, it does not require toxic methanol. Three such SAM- tutive promoters with different strengths (BBa_J23117, BBa_J23114, dependent methyltransferases have previously been identified in BBa_J23105, BBa_J23118, BBa_J23101, and BBa_J23100) (SI Ap- the plant maize (Zea mays) (8). Among them, the one called pendix, Table S4), or on the low-copy pTac15K vector under the anthranilic acid methyltransferase1 (AAMT1; encoded by aamt1) same set of synthetic promoters as well as the stronger isopropyl β was selected in this work as it was characterized to be the most -D-1-thiogalactopyranoside (IPTG)-inducible tac promoter (SI active for MANT formation in maize (8). Appendix, Fig. S5A). When these AAMT1 constructs, pTrc99A- Based on the selected SAM-dependent methylation-mediated series or pTac15K-series, were introduced to W3110 trpD9923 MANT synthesis pathway, we expressed AAMT1 in an ANT- strain and tested in two-phase flask culture, a clearly positive overproducing E. coli W3110 trpD9923 strain (9) (SI Appendix, correlation of MANT production with the strengths of the pro- Table S1) by introducing pTrcT that contained an E. coli codon- moters employed within the same vector series was observed, optimized version of aamt1 gene, designated as aamt1opt (SI Ap- while ANT accumulation was negatively correlated (SI Appendix, pendix,TableS2), under the control of trc promoter. The successful Fig. S5B). Also, it was found that under each of the six synthetic expression of AAMT1 in the recombinant W3110 trpD9923 strain promoters, AAMT1 expressed on the medium-copy pTrc99A harboring pTrcT was confirmed by SDS/PAGE (SI Appendix,Fig. vector enabled higher MANT production compared with those S2A). In a shake-flask culture using glucose as a sole carbon source, on the low-copy pTac15K (SI Appendix, Fig. S5B). Among these the W3110 trpD9923 strain harboring pTrcT successfully produced recombinant strains, W3110 trpD9923 strain harboring pTacT opt 35.8 ± 3.0 mg/L MANT (Fig. 2A and SI Appendix,Fig.S2B), along (aamt1 expressed on pTac15K under tac promoter) produced with the accumulation of 215.5 ± 7.5 mg/L ANT (SI Appendix,Fig. MANT to the highest titer of 297.3 ± 0.7 mg/L (Fig. 2A and SI S2B). No MANT production was detected in the wild-type W3110 Appendix, Fig. S5B), while accumulating ANT to the least trpD9923, although it produced a higher amount of ANT (305.4 ± amount of 117.1 ± 2.2 mg/L (SI Appendix, Fig. S5B). Based on opt 2.5 mg/L) (SI Appendix,Fig.S2B). these results, we further cloned aamt1 on pTac15K under T5 promoter with stronger strength than tac promoter (11), MANT Toxicity to E. coli. After the construction of a functional which however resulted in less MANT production (219.1 ± MANT biosynthesis pathway in E. coli, we next evaluated the 10.9 mg/L) (SI Appendix, Fig. S5B). toxicity of MANT to E. coli cells before further experiments. Exposing the wild-type E. coli W3110 cells to different concen- Increasing ANT Supply for MANT Production in E. coli. Next, we aimed trations of MANT (0.1, 0.2, 0.3, 0.5, 0.7, 0.8, and 1.0 g/L) in- at increasing intracellular flux to the direct precursor ANT to dicated a dose-dependent growth inhibition by MANT (SI further improve MANT production. ANT is a native metabolite in Appendix, Fig. S3). In the presence of 0.3 g/L MANT, the final L-TRP biosynthesis pathway in E. coli (Fig. 1A). Previous studies on optical density at the wavelength of 600 nm (OD600) was only a engineering E. coli for ANT overproduction have suggested several half of that obtained with the control experiment without MANT promising strategies for increasing ANT production (12, 13). First, exposure (SI Appendix, Fig. S3). When exposed in 1.0 g/L a feedback inhibition-resistant 3-deoxy-D-arabino-heptulosonate-7- fbr MANT, E. coli cell growth was completely inhibited (SI Appen- phosphate (DAHP) synthase (encoded by aroG )catalyzingthe dix, Fig. S3). condensation of erythrose 4-phosphate (E4P) and phosphoenolpyruvate (PEP) was expressed, by constructing pTrcGfbr (aroGfbr Designing an in Situ Two-Phase Extractive Culture Process. To cir- cloned on medium-copy pTrc99A under the strong trc promoter) cumvent this product toxicity limitation, a two-phase (aqueous/ and pBBR1Gfbr (aroGfbr cloned on low-copy pBBR1MCS under organic) cultivation system was designed, where MANT is the less strong lac promoter). W3110 trpD9923 strain harboring extracted in situ from the culture medium using an organic solvent pBBR1Gfbr produced 731.7 ± 7.4 mg/L ANT (SI Appendix,Fig. (Fig. 1). Tributyrin is one organic solvent nontoxic to microor- S6A), which was 2.3- and 1.4-fold higher than those obtained with ganisms including E. coli (10) and is also used in the food industry, W3110 trpD9923 strain harboring pTrcGfbr (SI Appendix,Fig.S6A) which can facilitate downstream purification processes for pre- and wild-type W3110 trpD9923 strain (SI Appendix,Fig.S2B), re- paring food-grade MANT. Tributyrin was observed to extract spectively. Second, we attempted to increase the availability of two MANT very efficiently as evidenced by its high partition co- key precursors E4P and PEP of the aromatic amino acid pathway efficient (420.1 ± 7.6) between aqueous medium phase and for improving ANT production. To increase E4P, the tktA gene tributyrin phase (SI Appendix, Table S3). Tributyrin could also encoding transketolase I was overexpressed by constructing extract ANT from the aqueous phase, but to a far lesser extent pBBR1Gfbr-A. W3110 trpD9923 strain harboring pBBR1Gfbr-A compared with that for MANT; the partition coefficients for produced 760.4 ± 12.5 mg/L ANT (SI Appendix,Fig.S6A), which ANT were 4.7 ± 0.2 and 0.1 ± 0.0 at pH 5.0 and 7.0, respectively was only slightly higher than that obtained with W3110 trpD9923 (SI Appendix, Table S3). The low partitioning of ANT into strain harboring pBBR1Gfbr. To increase PEP, several different tributyrin phase can be explained by its pKa value of 2.14. strains were constructed: ZWA1 (trc promoter exchange for ppsA

4of8 | www.pnas.org/cgi/doi/10.1073/pnas.1903875116 Luo et al. Downloaded by guest on October 2, 2021 overexpression), ZWA2 (pykF knockout), ZWA3 (pykF and pykA MET supplementation. This MANT titer corresponded to 12.7- double-knockout), and ZWA4 (trc promoter exchange for ppsA fold increase compared with that obtained with the initial W3110 overexpression and pykF knockout). However, flask cultures of trpD9923 strain harboring pTrcT in single-phase flask culture these four engineered strains each harboring pBBR1Gfbr-A showed without MET supplementation. This strain still accumulated no further increase in ANT titer (SI Appendix,Fig.S6A). The 346.7 ± 4.2 mg/L ANT. Also, ZWA4 harboring pBBR1GfbrAfbr fbr specific ANT production per OD600 of ZWA1, ZWA2, ZWA3, E and pTacTK produced a similarly high level of MANT ZWA4, and the parental strain (W3110 trpD9923)harboring (486.8 ± 4.8 mg/L) to that achieved with ZWA4 harboring pBBR1Gfbr-A was calculated to be 113.4, 114.6, 117.0, 112.0, and pBBR1GfbrAfbrEfbr and pTacT, but accumulated much less ANT −1 122.7 mg/L OD600 , respectively. Thus, ZWA4 strain was chosen (258.9 ± 7.6 mg/L) (SI Appendix, Fig. S8). Thus, these two strains for further engineering. Third, two additional overexpression targets were selected and further evaluated in fed-batch cultures. involved in L-TRP pathway (Fig. 1A), aroL (encoding shikimate kinase II) (14) and trpEfbr (encoding a feedback inhibition-resistant MANT Production in E. coli Two-Phase Fed-Batch Fermentations. Fed- ANT synthase) (15), were examined by constructing pTacLEfbr. batch cultures of the two engineered E. coli strains above were ZWA4 strain harboring pBBR1Gfbr-A and pTacLEfbr produced performed in two-phase mode for in situ extraction of MANT in a 820.5 ± 30.3 mg/L ANT (SI Appendix,Fig.S6A), which was 8.7% glucose minimal medium. The pH-stat feeding strategy was higher than that obtained with ZWA4 harboring pBBR1Gfbr-A. employed for nutrient feeding. Ammonium sulfate was added to The above-engineered strains overproducing ANT were provide additional nitrogen source. As it was found beneficial, combined with AAMT1 to produce MANT. To this end, three MET (20 mM) was supplemented to enhance SAM availability. combination strains were constructed: ZWA4 strain harboring Under these conditions, ZWA4 strain harboring pBBR1GfbrAfbr pBBR1Gfbr-A and LEfbr-pTacT, ZWA4 harboring pBBR1Gfbr-A Efbr and pTacTK produced 4.12 g/L MANT and 3.74 g/L ANT (SI and pTacT, and ZWA4 harboring pBBR1Gfbr and pTacT. Two- Appendix,Fig.S9), while ZWA4 harboring pBBR1GfbrAfbrEfbr phase flask cultures of these engineered strains showed that and pTacT produced 4.47 g/L MANT (Figs. 2 and 3A)and2.26g/L ZWA4 strain harboring pBBR1Gfbr-A and pTacT produced the ANT (Fig. 3A). For the latter strain, the yield and productivity of − − highest MANT titer (392.0 ± 1.7 mg/L) (Fig. 2A and SI Appendix, MANT were 0.045 g/g glucose and 0.062 g·L 1·h 1, respectively. Fig. S6B). Selecting C. glutamicum Chassis for Food-Grade MANT Production.

Enhancing SAM Availability for MANT Production in E. coli. Apart Having accomplished the proof-of-concept fermentative pro- SCIENCES from MANT production in the strains described above, ANT was duction of MANT from glucose by metabolically engineered E.

also formed as a byproduct. For example, ZWA4 strain harboring coli, we pursued to produce food-grade MANT by using other APPLIED BIOLOGICAL pBBR1Gfbr-A and pTacT, which so far produced the highest industrial GRAS microbial strains as MANT is mainly applied in MANT titer (392.0 ± 1.7 mg/L), accumulated 416.4 ± 7.8 mg/L food and cosmetic industries. Two such bacterial hosts, Pseudo- ANT at the same time (SI Appendix,Fig.S6B). Such a high-level monas putida KT2440 (Gram-negative) (17) and C. glutamicum accumulation of ANT indicated a potential bottleneck present in ATCC 13032 (Gram-positive) (18), were examined for their po- the conversion of ANT to MANT. One possible reason was tential to produce MANT by comparing their tolerance levels to thought to be limited availability of the cosubstrate SAM required MANT toxicity. The toxicity test showed that both strains had the by the AAMT1 reaction. The following strategies were applied to growth profiles exhibiting concentration-dependent inhibition by increase SAM availability by optimizing its biosynthesis (SI Ap- MANT (SI Appendix,Fig.S10), similar to E. coli.However,P. pendix,Fig.S7) and consequently to further increase MANT putida KT2440 cells could not grow in the presence of 1.0 g/L production while reducing ANT formation. First, the metAfbr gene MANT (SI Appendix,Fig.S10A). On the other hand, complete encoding feedback inhibition-resistant homoserine succinyl- growth inhibition of C. glutamicum was observed when 2.0 g/L fbr transferase and cysE encoding L-serine O-acetyltransferase (16) MANT was applied (SI Appendix,Fig.S10B). These results sug- were overexpressed for enhancing SAM biosynthesis (Fig. 1A). As gest that C. glutamicum has a higher tolerance to MANT com- a result, ZWA4 strain harboring pBBR1GfbrAfbrEfbr and pTacT pared with P. putida KT2440 and also E. coli.Thus,C. glutamicum produced MANT to a 6.8% higher concentration of 388.3 ± was selected as the host for food-grade MANT production (Fig. 6.0 mg/L and ANT to a 16.9% lower concentration of 346.3 ± 1B). Similar metabolic engineering strategies were applied to 3.3 mg/L (SI Appendix,Fig.S8) than those obtained with the pa- optimize MANT production in C. glutamicum (Fig. 2B). rental strain ZWA4 harboring pBBR1Gfbr and pTacT. Next, the metK gene encoding SAM synthetase was also overexpressed by Tuning AAMT1 Levels for MANT Production in C. glutamicum. To constructing pTacTK. ZWA4 strain harboring pBBR1Gfbr-A and construct MANT synthesis pathway, we expressed the aamt1opt pTacTK produced MANT to an 8.4% higher concentration of gene in the wild-type C. glutamicum ATCC 13032 strain and op- 424.8 ± 5.0 mg/L and ANT to a 16.4% lower concentration of timized its expression by constructing the following AAMT1 ex- 348.3 ± 7.8 mg/L (SI Appendix,Fig.S8) than those obtained with pression plasmids (SI Appendix,TableS5): pEKT (aamt1opt cloned ZWA4 harboring pBBR1Gfbr-A and pTacT. These three over- on pEKEx1 vector under tac promoter), pL10T, pI16T and pH36T expression targets were combined to construct ZWA4 harboring (aamt1opt cloned on pCES208 vector under L10, I16 and pBBR1GfbrAfbrEfbr and pTacTK, which produced ANT to the H36 promoters), pL10HT, pI16HT and pH36HT (aamt1opt cloned lowest concentration of 274.2 ± 8.4 mg/L. However, MANT titer on pCES208 vector under L10, I16, and H36 promoters with a N- was intriguingly decreased to 404.3 ± 5.4 mg/L compared with the terminal 6xHis-tag) (SI Appendix,Fig.S11A). These promoters (tac, best strain (ZWA4 harboring pBBR1Gfbr-A and pTacTK) so far L10, I16, and H36) have been reported to have different strengths of (SI Appendix,Fig.S8). Thus, L-methionine (MET), the direct gene expression (19), and the use of a His-tag was previously dem- precursor to SAM synthesis, was further supplemented to see onstrated to enhance gene expression in C. glutamicum (20). C. whether MANT production can be increased; this is thought to glutamicum ATCC 13032 strains harboring these plasmids were work because the metAfbr and cysEfbr genes we employed encode tested in single-phase shake-flask cultures supplemented with 0.8 g/L feedback inhibition-resistant enzymes in MET biosynthetic path- ANT. As a result, C. glutamicum ATCC 13032 strain harboring way. The three strains constructed above were cultured again in pH36HT produced MANT to the highest titer of 97.2 ± 8.6 mg/L (SI two-phase flasks with addition of 20 mM MET. As a result, the Appendix,Fig.S11B). No MANT was detected in the wild-type C. highest MANT titer of 489.0 ± 7.4 mg/L was obtained in glutamicum ATCC 13032 strain (SI Appendix,Fig.S11B). ZWA4 harboring pBBR1GfbrAfbrEfbr and pTacT (Fig. 2A and SI Plasmid pH36HT was thus introduced to C. glutamicum Appendix,Fig.S8), which was 25.9% higher than that without YTM1 strain (SI Appendix, Table S5), in which the trpD gene was

Luo et al. PNAS Latest Articles | 5of8 Downloaded by guest on October 2, 2021 Fig. 3. In situ two-phase extractive fed-batch culture profiles of (A) the engineered E. coli ZWA4 strain harboring pBBR1GfbrAfbrEfbr and pTacT and (B)the engineered C. glutamicum YTM8 strain harboring pSH36HTc and pEKGH under the condition of elevated DO level (50% air saturation). Symbols: blue circle,

cell growth (OD600); red rectangle, residual glucose concentration (in grams per liter); orange diamond, ANT concentration (in grams per liter); magenta triangle, MANT concentration (in grams per liter).

disrupted to block the L-TRP biosynthesis pathway in C. gluta- cultures, YTM3 strain produced 15.7% less ANT than did the micum after ANT (Fig. 1B). In flask culture without external ANT parental YTM2 strain; YTM3 strain harboring pEKG produced feeding and using glucose as a sole carbon source, YTM1 strain 4.8-fold more ANT than did YTM3 strain without pEKG; harboring pH36HT produced 45.3 ± 1.6 mg/L MANT (Fig. 2B YTM4 and YTM5 strains harboring pEKG produced similar and SI Appendix,Fig.S12)aswellas408.2± 4.2 mg/L ANT (SI levels of ANT to that obtained with the parental YTM3 strain Appendix,Fig.S12). Furthermore, we tested a C. glutamicum harboring pEKG (SI Appendix,Fig.S14A). These results collec- codon-optimized version of aamt1 gene, designated as aamt1opt-Cgl tively suggested that only aroGS180F overexpression by pEKG (SI Appendix, Table S6), by cloning it on pCES208 under could increase ANT production. Thus, aroGS180F was overex- H36 promoter with a N-terminal 6×His-tag in the same configu- pressed in the best MANT producer strain YTM2 harboring ration as for pH36HT, generating pH36HTc. Flask culture of pH36HTc. Since pH36HTc and pEKG harbor the same antibiotic YTM1 strain harboring pH36HTc produced MANT to a 1.6-fold (kanamycin) marker, we changed the antibiotic marker of higher concentration of 117.1 ± 2.0 mg/L (Fig. 2B and SI Ap- pH36HTc from kanamycin to spectinomycin/streptomycin, gen- pendix,Fig.S12) and ANT to a 55.0% lower concentration of erating pSH36HTc. Flask culture of YTM2 strain harboring 183.8 ± 3.1 mg/L (SI Appendix,Fig.S12), compared with those pSH36HTc produced 131.1 ± 9.0 mg/L MANT (SI Appendix, Fig. obtained by YTM1 strain harboring pH36HT. We further deleted S14B), which was almost the same as that obtained with the qsuB and qsuD genes in YTM1 strain, which are involved in YTM2 harboring pH36HTc (130.4 ± 0.8 mg/L). Plasmid pEKG competitive pathways toward protocatechuate and quinate syn- was then successfully introduced to YTM2 strain harboring thesis, respectively, in C. glutamicum (Fig. 1B), generating pSH36HTc, which produced MANT to a 91.6% higher concen- YTM2 strain. As a result, YTM2 harboring pH36HTc produced tration of 251.2 ± 2.5 mg/L (Fig. 2B and SI Appendix,Fig.S14B) MANT to an increased titer of 130.4 ± 0.8 mg/L (SI Appendix, Fig. and ANT to a 13.8-fold higher concentration of 3.38 ± 0.07 g/L (SI S12) compared with the parental strain (YTM1 harboring Appendix,Fig.S14B), compared with those obtained with pH36HTc). YTM2 strain harboring pSH36HTc alone. Up to this point, all flask cultures of C. glutamicum strains were conducted in single- Increasing ANT Supply for MANT Production in C. glutamicum. Next, phase cultivation mode. Thus, two-phase extractive flask culture of further metabolic engineering approaches were taken to increase YTM2 strain harboring pSH36HTc and pEKG was performed, the level of ANT and consequently to improve MANT production which produced MANT to a 44.9% higher concentration of in C. glutamicum. Although reported in E. coli (12, 13) and P. 364.1 ± 5.4 mg/L (Fig. 2B and SI Appendix,Fig.S14B) and ANT to putida (21), ANT overproduction has never been explored in C. a 21.0% lower concentration of 2.67 ± 0.03 g/L (SI Appendix, Fig. glutamicum. To increase ANT production in C. glutamicum,sev- S14B), compared with those obtained in single-phase culture. eraltargetgenesweremanipulated:e.g.,aroG, aroB,andaroK known as potential limiting steps in the shikimate pathway (Fig. Enhancing SAM Salvage for MANT Production in C. glutamicum. To 2B), and pgi, zwf, tkt, opcA, pgl,andtal known to enhance flux further improve MANT production in C. glutamicum,the through the pentose phosphate pathway (SI Appendix,Fig.S13), cosubstrate SAM availability was also considered. SAM over- which were target genes similarly manipulated for the overproduction in C. glutamicum has been reported in a previous study production of shikimic acid, 4-hydroxybenzoic acid, L-ornithine, (26), which focused on enhancing the flux to SAM biosynthesis and L-arginine in C. glutamicum (22–25). To this end, the fol- (Fig. 1B and SI Appendix,Fig.S15). Several effective strategies in lowing strains were constructed: YTM3 (with the native promoters that report were selected and applied for SAM production in this of aroK and aroB replaced with the strong constitutive sod pro- study. First, the metK gene encoding methionine adenosyl- moter in YTM2 strain), YTM4 (with the start codon of pgi transferase was overexpressed by constructing pEKGK. As a result, changed from ATG to GTG for down-regulation and of zwf YTM2 strain harboring pSH36HTc and pEKGK in two-phase flask changed from GTG to ATG for up-regulation in YTM3 strain), culture produced 377.0 ± 16.2 mg/L MANT (SI Appendix,Fig. YTM5 (with the native promoter of tkt replaced with sod pro- S16), which was only slightly higher than that obtained with moter in YTM4 strain), as well as plasmid pEKG (overexpressing YTM2 strain harboring pSH36HTc and pEKG. Next, the mcbR aroGS180F encoding a feedback-resistant DAHP synthase). In flask and Ncgl2640 genes encoding transcriptional regulators involved in

6of8 | www.pnas.org/cgi/doi/10.1073/pnas.1903875116 Luo et al. Downloaded by guest on October 2, 2021 regulation of MET biosynthesis were deleted in YTM2 strain, ANT Production. Although it is not the major subject of this study, generating YTM6. Furthermore, the metB gene encoding cys- ANT is also an important industrial platform chemical with many tathionine-γ-synthase involved in competitive synthesis of L- applications (21), and its biobased production is an active re- isoleucine was deleted in YTM6 strain, generating YTM7. However, search subject (12, 13, 21). During the two-phase extractive two-phase flask cultures of YTM6 and YTM7 strains harboring fermentation at neutral pH performed in this study, it is no- pSH36HTcandpEKGshowedseverely retarded cell growth, and ticeable that all of the MANT was extracted into the tributyrin consequently produced 80.0 ± 4.2 and 72.3 ± 5.5 mg/L MANT, phase, while almost all ANT (e.g., 96.9% in the case of fed-batch fbr fbr fbr respectively (SI Appendix,Fig.S16), which were far lower than that culture of E. coli ZWA4 harboring pBBR1G A E and obtained with YTM2 strain harboring pSH36HTc and pEKG. In pTacT, 98.4% in the case of fed-batch culture of C. glutamicum addition to genetic engineering for SAM synthesis, we directly YTM8 harboring pSH36HTc and pEKGH) was retained in supplemented the SAM precursor MET (10 mM) in two-phase aqueous phase. This is particularly beneficial for downstream flask culture of YTM2 strain harboring pSH36HTc and pEKG as processes to purify MANT and also ANT from the fermentation it had increased MANT titer in E. coli. Unfortunately, the direct broth. From a process engineering perspective, this could be supplementation of MET resulted in the production of 326.8 ± regarded as a coproduction of both MANT and ANT. To explore 3.9 mg/L MANT (SI Appendix,Fig.S16), lower than that without the capacity for sole ANT production, we further performed a MET supplementation. From these results, it was speculated that single-phase fed-batch culture using one of the engineered ANT SAM availability might not be a limiting factor for MANT pro- overproducers developed in this study, i.e., C. glutamicum duction in C. glutamicum. YTM5 strain harboring pEKG. The single-phase fermentation of Thus, a different strategy of engineering the SAM salvage this strain led to the production of 26.40 g/L ANT at 84 h in pathway (Fig. 1B) was applied for recycling the SAM reaction glucose minimal medium (SI Appendix, Fig. S21), representing not only the production of this compound by C. glutamicum, but product S-ribosyl-L-homocysteine (SAH) back to SAM synthesis pathway catalyzed by SAH hydrolase (encoded by sahH) (27). The also the highest ANT production titer reported in any microbial sahH gene was overexpressed from pEKGH. To our pleasant host to date. On the other hand, the residual amount of this surprise, two-phase flask culture of YTM2 strain harboring precursor metabolite indicates that there is still some room for pSH36HTc and pEKGH produced MANT to a 63.9% higher improvement of MANT production. One promising strategy would be to engineer the ANT methyltransferase by enzyme concentration of 596.9 ± 16.7 mg/L (Fig. 2B and SI Appendix,Fig.

evolution for greater catalytic activity, thereby more efficiently SCIENCES S16) and ANT to a 5.2% lower concentration of 2.53 ± 0.01 g/L converting the residual ANT to MANT. (SI Appendix,Fig.S16), compared with those obtained with In summary, we report the development of metabolically APPLIED BIOLOGICAL YTM2 strain harboring pSH36HTc and pEKG. This MANT titer engineered E. coli and C. glutamicum strains capable of pro- is 12.3-fold higher than that obtained with the initial YTM1 strain ducing MANT by direct fermentation in minimal media con- harboring pH36HT in single-phase flask culture. taining glucose as a sole carbon source. A synthetic metabolic pathway for MANT biosynthesis was constructed in both E. coli MANT Production in C. glutamicum Two-Phase Fed-Batch Fermentations. and C. glutamicum, and MANT production was optimized by Since C. glutamicum showed higher tolerance to MANT toxicity, tuning the key enzyme AAMT1 expression, increasing flux to- we first performed a single-phase fed-batch culture using the ward precursor ANT, and increasing availability (in E. coli) and best C. glutamicum MANT producer strain YTM2 harboring salvage (C. glutamicum) of cosubstrate SAM required by the pSH36HTc and pEKGH in glucose minimal medium without AAMT1 reaction. In situ two-phase extractive fed-batch fer- MET supplementation, which led to production of 1.70 g/L MANT mentation process was also developed for MANT production by and 14.11 g/L ANT (SI Appendix,Fig.S17).Next,insitutwo-phase the engineered E. coli and C. glutamicum, which led to pro- extractive fed-batch culture of YTM2 strain harboring pSH36HTc duction of 4.47 and 5.74 g/L MANT from glucose, respectively. and pEKGH under the same settings was performed, which pro- These titers are significantly high for natural compounds produced MANT to a 1.36-fold higher concentration of 4.01 g/L and duced by microbial fermentation, as most natural compounds ANT to an 86.1% lower concentration of 1.96 g/L (SI Appendix, from engineered microbes are produced at levels of milligrams Fig. S18), compared with those obtained in single-phase fermen- or micrograms per liter (28). Also, all fermentations described in tation. In this two-phase fermentation, succinic acid was also ac- this study were performed in minimal media, which further help cumulated to 10.06 g/L. Compared with single-phase fermentation, decrease operation and separation costs (29). This methanol-free the two-phase fermentation resulted in an emulsion-like environ- biobased production of MANT paves the way toward a sustain- ment, which might interfere with oxygen transferred to the cells. It able, consumer-friendly process to produce MANT as a “natu- was hypothesized that the high succinic acid accumulation was due ral” flavoring agent, which for the past 100 y has only been to oxygen limitation. Thus, the dissolved oxygen (DO) concentra- produced industrially through chemical synthesis. The metabolic tion was increased by setting the control DO value from 30 to 50% engineering strategies and methodologies described and the of air saturation and also by changing agitation speed from the engineered microbial systems established here will contribute to fixed at 600 rpm to automatic increase from 600 to 1,000 rpm the development of engineered strains for the sustainable pro- according to DO. By increasing the DO level, the two-phase fed- duction of other chemicals with similar properties and charac- batch culture of YTM2 strain harboring pSH36HTc and pEKGH teristics as MANT. produced MANT to a 30.9% higher concentration of 5.25 g/L at the 110-h mark and ANT to a 2.0-fold higher concentration of Materials and Methods 5.90 g/L (SI Appendix,Fig.S19), with succinic acid reduced to All of the materials and methods conducted in this study are detailed in SI 5.09 g/L. Glycerol formation was also observed in the fermentation Appendix, Supplementary Information Materials and Methods, including broth, and thus the hdpA gene involved in glycerol formation bacterial strains and media, construction of E. coli expression plasmids, E. coli pathway was also deleted (SI Appendix,Fig.S20) (22, 23) in genome manipulation, construction of C. glutamicum expression plasmids, YTM2 strain, generating YTM8. As a result, two-phase fed-batch construction of CRISPR-based recombineering plasmids for gene knockout in culture of YTM8 strain harboring pSH36HTc and pEKGH pro- C. glutamicum, construction of genetic engineering plasmids for promoter exchange and for further gene knockout in C. glutamicum YTM2, C. gluta- duced MANT to a 9.3% higher concentration of 5.74 g/L at 110 h micum genome manipulation, MANT toxicity test, SDS/PAGE analysis, de- (Figs. 2B and 3B) and ANT to a 33.7% higher concentration of termination of partition coefficients, cultivation condition, and analytical 7.89 g/L (Fig. 3B). The resultant yield and productivity of MANT procedures. The data supporting the findings of this study are available in −1 −1 were 0.020 g/g glucose and 0.052 g·L ·h , respectively. SI Appendix.

Luo et al. PNAS Latest Articles | 7of8 Downloaded by guest on October 2, 2021 ACKNOWLEDGMENTS. We thank Prof. Ki Jun Jeong for generously pro- to Solve Climate Changes on Systems Metabolic Engineering for Biorefi- viding us with plasmids pCES-L10-M18, pCES-I16-M18, and pCES-H36-M18 for neries from the Ministry of Science and ICT through the National Research cloning in this study. We also thank Tae Hee Han for providing plasmid Foundation (NRF) of Korea (Grants NRF-2012M1A2A2026556 and NRF- pSY06b. This work was supported by the Technology Development Program 2012M1A2A2026557).

1. Wang J, De Luca V (2005) The biosynthesis and regulation of biosynthesis of Concord 16. Kunjapur AM, Hyun JC, Prather KLJ (2016) Deregulation of S-adenosylmethionine grape fruit esters, including “foxy” methylanthranilate. Plant J 44:606–619. biosynthesis and regeneration improves methylation in the E. coli de novo vanillin 2. Yadav GD, Krishnan MS (1998) An ecofriendly catalytic route for the preparation of biosynthesis pathway. Microb Cell Fact 15:61. — perfumery grade methyl anthranilate from anthranilic acid and methanol. Org Pro- 17. Loeschcke A, Thies S (2015) Pseudomonas putida a versatile host for the production – cess Res Dev 2:86–95. of natural products. Appl Microbiol Biotechnol 99:6197 6214. — 3. Walton NJ, Mayer MJ, Narbad A (2003) Vanillin. Phytochemistry 63:505–515. 18. Becker J, Wittmann C (2012) Bio-based production of chemicals, materials and fuels – 4. Kittleson JR, Pantaleone DP (1994) US Patent 5437991. Corynebacterium glutamicum as versatile cell factory. Curr Opin Biotechnol 23:631 5. Taupp M, Harmsen D, Heckel F, Schreier P (2005) Production of natural methyl an- 640. 19. Yim SS, An SJ, Kang M, Lee J, Jeong KJ (2013) Isolation of fully synthetic promoters for thranilate by microbial N-demethylation of N-methyl methyl anthranilate by the high-level gene expression in Corynebacterium glutamicum. Biotechnol Bioeng 110: topsoil-isolated bacterium Bacillus megaterium. J Agric Food Chem 53:9586–9589. 2959–2969. 6. Berger RG, Drawert F, Hadrich S (1988) Microbial sources of flavour compounds. In 20. Shin JH, et al. (2016) Metabolic engineering of Corynebacterium glutamicum for ’ Bioflavour 87, ed Schreier P (Walter de Gruyter, Berlin). enhanced production of 5-aminovaleric acid. Microb Cell Fact 15:174. 7. Gross B, et al. (1990) Production of methylanthranilate by the basidiomycete Pycno- 21. Kuepper J, et al. (2015) Metabolic engineering of Pseudomonas putida KT2440 to – porus cinnabarinus (Karst.). Appl Microbiol Biotechnol 34:387 391. produce anthranilate from glucose. Front Microbiol 6:1310. 8. Köllner TG, et al. (2010) Herbivore-induced SABATH methyltransferases of maize that 22. Kogure T, Kubota T, Suda M, Hiraga K, Inui M (2016) Metabolic engineering of Co- methylate anthranilic acid using S-adenosyl-L-methionine. Plant Physiol 153:1795– rynebacterium glutamicum for shikimate overproduction by growth-arrested cell 1807. reaction. Metab Eng 38:204–216. 9. Yanofsky C, Horn V, Bonner M, Stasiowski S (1971) Polarity and enzyme functions in 23. Kitade Y, Hashimoto R, Suda M, Hiraga K, Inui M (2018) Production of 4-hydrox- mutants of the first three genes of the tryptophan operon of Escherichia coli. Genetics ybenzoic acid by an aerobic growth-arrested bioprocess using metabolically en- 69:409–433. gineered Corynebacterium glutamicum. Appl Environ Microbiol 84:e02587-17. 10. Kim B, Park H, Na D, Lee SY (2014) Metabolic engineering of Escherichia coli for the 24. Kim SY, Lee J, Lee SY (2015) Metabolic engineering of Corynebacterium glutamicum – production of phenol from glucose. Biotechnol J 9:621–629. for the production of L-ornithine. Biotechnol Bioeng 112:416 421. 11. Ajikumar PK, et al. (2010) Isoprenoid pathway optimization for Taxol precursor 25. Park SH, et al. (2014) Metabolic engineering of Corynebacterium glutamicum for L- overproduction in Escherichia coli. Science 330:70–74. arginine production. Nat Commun 5:4618. 12. Balderas-Hernández VE, et al. (2009) Metabolic engineering for improving anthra- 26. Han G, Hu X, Qin T, Li Y, Wang X (2016) Metabolic engineering of Corynebacterium glutamicum ATCC13032 to produce S-adenosyl-L-methionine. Enzyme Microb Technol nilate synthesis from glucose in Escherichia coli. Microb Cell Fact 8:19. 83:14–21. 13. Noda S, Shirai T, Oyama S, Kondo A (2016) Metabolic design of a platform Escherichia 27. Lozada-Ramírez JD, Martínez-Martínez I, Sánchez-Ferrer A, García-Carmona F (2008) coli strain producing various chorismate derivatives. Metab Eng 33:119–129. S-Adenosylhomocysteine hydrolase from Corynebacterium glutamicum: Cloning, 14. Dell KA, Frost JW (1993) Identification and removal of impediments to biocatalytic overexpression, purification, and biochemical characterization. J Mol Microbiol Bio- synthesis of aromatics from D-glucose: Rate-limiting enzymes in the common pathway technol 15:277–286. of aromatic amino acid biosynthesis. J Am Chem Soc 115:11581–11589. 28. Park SY, Yang D, Ha SH, Lee SY (2018) Metabolic engineering of microorganisms for 15. Gu P, Yang F, Kang J, Wang Q, Qi Q (2012) One-step of tryptophan attenuator in- the production of natural compounds. Adv Biosyst 2:1700190. activation and promoter swapping to improve the production of L-tryptophan in 29. Lee SY, Kim HU (2015) Systems strategies for developing industrial microbial strains. Escherichia coli. Microb Cell Fact 11:30. Nat Biotechnol 33:1061–1072.

8of8 | www.pnas.org/cgi/doi/10.1073/pnas.1903875116 Luo et al. Downloaded by guest on October 2, 2021