Characterization of Highly Reductive Modication of Tetracycline D-Ring Reveals Enzymatic Conversion of Enone to Alkane

Tetracyclines are an eminent family of type II polyketides which possess a variety of decoration on the skeletons. However, apart from the oxidative modication in aureolic acid compounds, there are few cases on the further conversion of α, β-unsaturated ketones in the tetracycline D-ring. Here, we identied two reductases (TjhO5 and TjhD4) that highly reduced the α,β-unsaturated ketone of D-ring in the biosynthesis of unconventional tetracyclines. By identifying related intermediates and conducting isotope incorporation experiments, we demonstrated that the entire transformation could be accomplished by TjhO5 and TjhD4 collectively via two distinct pathways involving different enzymatic mechanisms. A distinctive deoxygenation mechanism was possibly involved in the TjhO5-mediated continuous reduction of C = O to CH 2 . These ndings highlight the unprecedent post-modication of tetracyclines and facilitate further engineering to enrich the structural diversities.


Introduction
Type II polyketides belong to a structurally diverse family of natural products with various biological activities, [1][2][3] and are closely related to the human microbiome. 4 As an essential class of type II polyketides, tetracyclines are clinically used to treat a variety of infections. 5,6 During the biosynthesis of bacterial tetracyclines, such as tetracycline (TC), chlortetracycline (CTC), oxytetracycline (OTC), and SF2575, 7-10 the same intermediate 2,4-keto-anhydrotetracycline (ATC) can be further modi ed by various tailoring enzymes to obtain diverse tetracyclines ( Fig. 1a and Supplementary Fig. 1a) [11][12][13][14] . Nevertheless, the α,β-unsaturated ketone of D-ring is generally retained in the nal products except aureolic acid compounds including mithramycin [15][16][17][18] , whose scaffolds are formed via cleaving the D-ring of tetracycline precursors through Baeyer-Villiger oxidation ( Supplementary Fig. 1b). [19][20][21][22] In recent study, we activated and characterized a silent gene cluster tjh in Streptomyces aureus suzhoueusis governing the biosynthesis of two types of type II polyketides (pentacyclic and tetracycline structures). 23 Compared with typical tetracyclines containing α,β-unsaturated ketones, the D-rings of these non-canonical tetracyclines have been highly reduced to either C = C (5-7) or C-C bond (9-11) ( Fig. 1a). Considering that three new compounds (12)(13)(14) with similar core structure to 4-hydroxy-ATC have been previously identi ed in ΔtjhO5 mutants, the atypical tetracyclines are likely to be converted from intermediates with α,β-unsaturated ketone ( Fig. 1a and Supplementary Fig. 1a). Intuitively, the transformation of enone to alkane in D-ring can be divided into two sub-processes consisting of reduction of the C = C bond and conversion of C = O to CH 2 . The strategies for reducing ketone to alkane in organic synthesis have been widely developed mainly including Wolff-Kishner-Huang reduction, Clemmensen reduction and Caglioti reduction (Fig. 1b). 24 Correspondingly, such conversion in nature mainly occurs in type I polyketides biosynthesis, which is achieved through three enzymatic domains of type I polyketide synthase (PKS) including ketoreductase (KR), dehydratase (DH) and enol reductase (ER) (Fig. 1c). 25 In this study, we found that the quinone oxidoreductase TjhO5 can mediate this continuously reductive reaction of C = O to CH 2 in vivo involving a unique deoxygenation mechanism. In addition, TjhO5 along with a NAD(P)H-dependent epimerase TjhD4 accomplished the conversion of enone to alkane in D-ring of tetracyclines. Signi cantly, this process was veri ed to go through two different biosynthetic pathways, which involved distinct intermediates and enzymatic mechanisms based on isotope labeling experiments.

Results
Biochemical characterization of TjhO5 and identi cation of the product. According to BLAST analysis, TjhO5 belongs to the medium-chain reductases (MDR) superfamily and NADPH:quinone oxidoreductase subfamily; it exhibits a high similarity to GrhO7 (50% identity) whose function in griseorhodin biosynthesis is not assigned. 26,27 Since 12-14 can deglycosylate spontaneously and 8 (with a reduced completely D-ring) can be detected in ΔtjhB3 (tjhB3 encoding a glycosyltransferase), the deglycosylated compound 15 should be more reliable and suitable for further studies. We generated substantial amounts of 15 via hydrolysis of ΔtjhO5-fermentation broth with hydrochloric acid and con rmed its structure by analysis of NMR spectroscopy data . Moreover, retention time (RT) of 15 was identical to that of the deglycosylated compound in the crude extract of the double knock-out mutant ΔtjhO5/ΔtjhB3 (Supplementary Fig. 2-3). To pinpoint its function in vitro, we puri ed TjhO5 from E. coli BL21(DE3) (Supplementary Fig. 4). Different from the control assays, LC − MS showed two new peaks whose m/z ([M-H] -= 369) was consistent with the target product 4 in the reaction mixture containing NADPH, substrate 15 and TjhO5. Surprisingly, the RT of the two peaks was different from that of 4, the major core structure produced by wild-type (WT) (Fig. 2a).
To investigate TjhO5-catalyzed reaction, the major product 16 was puri ed via a large-scale enzymatic assay of TjhO5. Further, the structure of 16 was con rmed by 1 H, 13 [24][25][26][27][28][29]. The structure of compound 16 is similar to that of 15, except that its C-10 carbonyl group was reduced to CH 2, and the C-7 of 16 is the R-con guration instead of the S-con guration in 15 (Fig. 2b). These differences were revealed by the evaluation of the H-7 and H-6a coupling constants in 1 H NMR data of 15 and 16, which were 11.1 Hz and 1.6 Hz, respectively ( Fig. 2c and d). Besides, in NOESY NMR spectrums, H-7 of 16 was correlated with H β -6, but H-7 of 15 was correlated with H α -6 ( Supplementary Fig. 5). Meanwhile, a minor product (marked with * in Fig. 2a-i) was not obtained due to its low yield and lability.
In vitro reconstitution of highly reductive modi cation. At this stage, it is obvious that other enzymes should participate in further transformation to generate nal products starting from the α,β-unsaturated ketone intermediate. Besides tjhO5, gene knock-out experiments veri ed three genes, including tjhC5, tjhD2, and tjhD4, to be closely related to the post-modi cations of tetracyclines. 23 Based on bioinformatic analysis, TjhC5 contains a cyclase domain; TjhD2 is an aldo-keto reductase, sharing moderate homology (33% amino acid identity) with SsfF which can reduce the C-4 carbonyl of 2,4-keto-ATC; TjhD4 belongs to the short-chain dehydrogenase (SDR) superfamily and NAD + dependent epimerase/dehydratase subfamily, which is most likely to participate in subsequent reduction. To verify this hypothesis, we performed various biochemical assays on TjhD4 from E. coli BL21(DE3) (Supplementary Fig. 4). When intermediate 16 was incubated with TjhD4 and NAD(P)H for 10 min and LC-MS was performed on products, two new products with the same m/z ([M-H] -= 371) were detected. The major product was identi ed as 8 and the other was a new compound 17 ( Fig. 3a i-iii). The chemical structure of 17 was elucidated by NMR spectroscopy analysis , and its C-9 is R-con guration rather than S-con guration in 8 (Fig. 3b). In fact, 17 could be detected in the fermentation broth of WT but was ignored previously due to low titer. Although TjhD4 could convert 16 to 8 and 17 in vitro, both were minor backbones produced by WT and could not be further converted into the predominant scaffold 4. Also, the extract of TjhO5 reactions as substrates for TjhD4 could not afford 4, suggesting that products of TjhO5 alone could not be transformed to 4 by TjhD4 ( Supplementary Fig. 6). Accidentally, incubation of 15 with TjhO5 and TjhD4 at the same time generated trace amounts of 4. Furthermore, adjusting the ratio of TjhO5 and TjhD4 from 3:1 to 1:3 under conditions of excess substrate 15 signi cantly improved the yield of 4 ( Fig. 3a v-vii). Consequently, we proposed that the formation of 4 needed an optimal ration of both enzymes because excess TjhO5 might convert 4 to 8 and 17 (Fig. 3b). As expected, incubation of TjhO5, 4 and NADPH resulted in the appearance of 8 and 17, and concomitant disappearance of 4 ( Fig. 3a x-xi). These results suggested that i) either TjhO5 or TjhD4 was indispensable for transformation of the α,β-unsaturated ketone substrate 15 to 4, 8, and 17; ii) TjhO5, a multifunctional enzyme, could not only reduce the ketone to CH 2 and isomerize the C-7 but also reduce different functional groups including C = O and C = C double bonds; iii) TjhD4 is a bi-functional enzyme which could catalyze epimerization and reduction reactions simultaneously; iv) the formation of 8 and 17 beginning with 15 went through two different intermediates respectively via two distinct biosynthetic pathways (path A and B) which appeared in the stepwise and one-pot reaction, respectively (Fig. 3b).
Elucidation of the enzymatic mechanisms in two distinct pathways. For an in-depth analysis of the enzymatic mechanism, the activity of TjhO5 was reconstituted in the presence of (R)-[4-2 H]-NADPH generated in situ by glucose dehydrogenase (GDH) from Thermoplasma acidophilum ATCC 25905 (TaGDH), using D-[1-2 H]-glucose as the deuterium donor. 28,29 The mass shift of + 2 Da at m/z 371 was detected in a TaGDH/TjhO5 coupled assay ( Supplementary Fig. 7). We prepared 2 H-16 via large-scale enzymatic reaction and 2 H NMR revealed that two deuteriums (δ 2 H 2.48, δ 2 H 2.70) were incorporated in 16 at C-10 ( Fig. 4a and b). Meanwhile, enzymatic assay of TjhD4 was conducted in the presence of (S)-[4-2 H]-NADPH provided by a GDH from Bacillus megaterium DSM 2894 (BmGDH) using the same deuterium donor. 30 BmGDH/TjhD4 coupled assay could produce two compounds, both showing a mass increment of + 1 Da (Supplementary Fig. 8).
Given that the con guration of C-7 was reversed in TjhO5-governed reaction in path A, we proposed that there were two plausible catalytic mechanisms for forming 16. TjhO5 rstly reduced the carbonyl of C-10 to hydroxy resulting in compound 18 ( Supplementary Fig. 11a). In one of the hypotheses, 18 was then dehydrated and tautomerized to form the intermediate 20, which could be transformed to 16 by 1,4addition. Alternatively, 19 could be reduced to intermediate 21, which subsequently went through four times of enol interconversions to form 16. To con rm the mechanism, we introduced enzymatic reactions in D 2 O at rst. The mass data of 16 did not change revealed by LC-MS, suggesting the reaction did not abstract the proton from water to C-7 ( Supplementary Fig. 11b). Therefore, these mechanisms which require the incorporation of a solvent-derived proton into products were invalidated. 31,32 Since the oxygen of carbonyl in β-hydroxyl-α,β-enone was easily exchanged with water oxygen, 33 Fig. 5b and Supplementary Fig. 13). These results suggested that oxygen of C-10 was retained in this reaction and might migrate to C-7. We proposed that the deoxygenation was possibly achieved via generating a potential cyclic ether intermediate in the virtue of acid residues (Fig. 5a), reminiscent of acid-mediated cyclic ether formation in platensimycin biosynthesis. 34 After completion of the TjhO5 reaction in H 2 18 O, TjhD4 was added to afford 8 and 17, whose MW still increased by + 2 and + 4 ( Supplementary Fig. 14).
However, when simultaneously incubating TjhO5, TjhD4, NADPH and 15 in H 2 18 O, we detected only + 2 increase in MW of all products, suggesting that path A and path B involved different enzymatic mechanisms ( Fig. 3b and Supplementary Fig. 14).  Fig. 15). These results veri ed that the formation of 4 did not require oxygen transfer, as well as abstraction of a proton from water to carbon. Therefore, in the co-incubation of TjhD4 and TjhO5, Hprovided by TjhD4 rapidly attacked 18 to afford 4 ( Supplementary Fig. 15), which could be further reduced by TjhO5 to yield 8 and 17.

Discussion
We successfully reconstituted the highly reductive modi cation of tetracycline D-ring by TjhO5 and TjhD4 in vitro, involving two different biosynthetic pathways with distinct enzymatic mechanisms (Fig. 6). In path A, TjhO5 reduces the carbonyl group of C-10 to hydroxy to form the potential intermediate 22, which can be reduced by TjhO5 again to produce 16. TjhD4 could epimerize C-7 and reduce C-8 to form 8 and 17. The enzymatic function of reducing double bond or deoxygenation has been well studied.
Signi cantly, it is rarely reported that one reductase is responsible for the conversion of C = O to CH 2 , because such transformation usually involves multi-step reduction and deoxygenation. More interestingly, the C-10 deoxygenation characterized in this study is inherently mechanistically different from dehydration catalyzed by KstA10 in kosinostatin biosynthesis 35 , radical mechanism in apramycin, IPP and DMAPP biosynthesis 36,37 , and α-carbonyl mediated mechanism guided by the PMP-dependent enzyme SpnQ in D-forosamine biosynthesis 38 (Supplementary Fig. 16a-d). TjhO5 may remove a hydroxy of C-10 via the oxygen transfer mechanism, accompanying hydroxy isomerization. Moreover, TjhD4 belongs to the NADH-dependent epimerases, which usually participate in deoxysugar biosynthesis and exploit the enol interconversion of α-carbonyl to isomerize (Supplementary Fig. 16e). 39,40 Although the catalytic mechanism of homologous enzymes has been well elucidated, to our knowledge, it has not been reported that this enzyme can conduct reduction and isomerization on the backbone of type II polyketides. In path B, two enzymes are required to work coordinately in the rst step. After TjhO5 reduces the carbonyl of C-10, TjhD4 rapidly catalyzes the attack of Hat C-8 of 18, resulting in the disappearance of hydroxy to produce 4, which is reduced to 8 and 17 by TjhO5 again (Fig. 6). TjhO5 may exhibit a jumping catalytic mode in this pathway rather than a continuous catalytic mode in path A. In addition, the phenomenon that the same products are formed through two different pathways with distinct intermediates from a common substrate is rare. 41 Consequently, TjhO5 and TjhD4 both exhibit high catalytic promiscuity in these two pathways.
The sequence similarity network (SSN) analysis of TjhO5 revealed that the group closely related to TjhO5 is different from most quinone oxidoreductases ( Supplementary Fig. 17). More speci cally, proteins with high homology to TjhO5 are extensively involved in the biosynthesis of type II polyketides, such as GrhO7, MsnO9 and HrsP2, which are related to the biosynthesis of griseorhodin, mensacarcin and hiroshidine respectively ( Supplementary Fig. 18). 26,42,43 Although quinone oxidoreductases are commonly involved in type II polyketides biosynthesis, the functions of most of these proteins remains to be fully determined.
This study lays the foundation for functional characterization of these enzymes.
In summary, we identi ed two enzymes that could cooperatively reduce a D-ring with α,β-unsaturated ketone to a saturated D-ring via two distinct pathways with different enzymatic mechanisms. This study highlights the reactive diversity of reductases and sheds light on the role of quinone oxidoreductases and epimerases in the biosynthesis of type II polyketides, which can be applied in genome mining, biocatalysis and combinatorial biosynthesis.

Methods
General. Speci c bacterial strains and plasmids used in this study were summarized in Table S1, PCR primers were listed in Table S2. General enzymes, chemicals, Kits, media, and molecular biological reagents were from standard commercial sources. Bioinformatics analysis, DNA isolation, manipulation, construction of gene replacement and complementation mutants were preformed following the standard methods.
Hydrolysis of fermentation broth. The fermentation broth of ΔtjhO5::2R (TG6027) was dissolved by suitable amount of methanol with 0.25 M hydrochloric acid at 25 ℃ stirring for 1-3 h.
Enzymatic assays and metabolite analysis. High performance liquid chromatography (HPLC) analysis was conducted on Thermo Scienti c Dionex Ultimate 3000 (Thermo Fisher Scienti c Inc., USA) with a reverse-phase Alltima C18 column (5 µm, 4.6×250 mm). Semi-preparative HPLC was performed on a Shimadzu LC-20-AT system using an YMC-Pack ODS column (YMC, 250×10 mm, 5 µm). Protein Expression and puri cation. The TjhO5 gene was ampli ed by PCR from genome DNA using the primers shown in Table S2. The puri ed PCR products were ligated to pMD19-T and con rmed by sequencing. The code of TjhD4 was optimized and this gene was synthesized and cloned into pUC57 by Genewiz. The NdeI/HindIII fragment was cloned into the same sites of pET28a to yield plasmid pTG6032 (TjhO5) and pTG6033 (TjhD4), in which enzyme will be overproduced as a C-terminal 8x His-tagged fusion protein. For TjhO5 and TjhD4 expression and puri catin, plasmid pTG6032 and pTG6033 was transformed respectively into BL21(DE3) competent Escherichia coli cells which were grown at 37 ℃ in 800 mL LB with 50 µg/mL kanamycin to an OD600 of 0.6-0. Enzymatic assay.