Geldanamycin (1, Figure 1), the first member of benzoquinone ansamycins, was isolated from Streptomyces hygroscopicus in 1970.1 Although exhibiting potent cytotoxicity against various cancer cells, 1 is not a clinical compound due to its severe hepatotoxicity and poor water solubility.2 17-AAG (17-allylamino-17-demethoxygeldanamycin) as a semisynthetic derivative of 1 with much improved water solubility is currently under clinical trial for breast cancer treatment.3 Many new analogs or derivatives of 1 have been created or discovered in the past few years.4, 5, 6, 7, 8

Figure 1
figure 1

Chemical structure of 1, 2 and KOSN1631. A full color version of this figure is available at The Journal of Antibiotics journal online.

We are interested in natural 1 analogs and understanding their synthetic mechanisms. We identified such analogs as 4,5-dihydro-4-hydroxygeldanamycins, thiazinogeldanamycin and 19-S-methylgeldanamycin from S. hygroscopicus 17997 and characterized their synthetic mechanisms.9, 10, 11, 12, 13 We also discovered a minor component 7-descarbamoyl-7-hydroxygeldanamycin from a gdmN disruption mutant of S. hygroscopicus 17997, which presented an additional proof for C-7 carbamoylation taking place before C-4,5 oxidation in 1 biosynthesis.14

Recently, as a result of our continued efforts for natural 1 analogs, we discovered 6-demethoxy-6-methylgeldanamycin (2) in 1 preparation from S. hygroscopicus 17997. In this paper, we reported the structure of 2 and its implication of 1 biosynthesis.

Some preparations of 1 were found to contain 1 analogs as small or trace impurities.15, 16 The HPLC of our 1 preparation (with a purity of about 90%; see Supplementary material: a brief description of 1 preparation from S. hygroscopicus 17997) from S. hygroscopicus 17997 displayed a small peak at 24.7 min (about 1.7% of the principle 1 peak at 22.3 min; Figure 2). The peak revealed a molecular ion at m/z 567 ([M+Na]+), which exhibited a typical MS2 fragment pattern of 1 (Supplementary Figure S1). The m/z 567 aroused our interests, as we could not assign a reasonable structure for it from MS data and current understanding of 1 biosynthesis.17, 18, 19

Figure 2
figure 2

HPLC of geldanamycin (1) preparation from S. hygroscopicus 17997. HPLC parameters: Agilent 1200 RRLC system (Agilent, Waldbronn, Germany); Dikma Diamonsil C18 column (4.6 × 150 mm, 5 μm, DIKMA, Beijing, China), mobile phase MeOH-H2O, 40–100% in 30 min, 1.0 ml min−1, wavelength 304 nm. A full color version of this figure is available at The Journal of Antibiotics journal online.

To elucidate the structure of the analog with m/z 567 (2) by NMR, a total amount of 1070 mg 1 preparation, dissolved in 10 ml dimethyl sulfoxide, was used to make a pure preparation of 2 by reversed-phase HPLC (Shimadzu LC-20AP, SHIMADZU, Kyoto, Japan; YMC ODS-A, 21.2 × 150 mm, mobile phase MeOH-H2O, 62–100% in 21 min, 12.5 ml min-1, wavelength 254 nm; Supplementary Figure S2). After evaporation, an amount of 5.6 mg of 2 as yellow amorphous powder was obtained. Analytical HPLC indicated that it displayed an UV absorption profile very similar to that of 1 (Supplementary Figure S3).

The molecular formula of 2 was established as C29H40N2O8 by HR-ESI(+)-MS (m/z 567.26596, calculated 567.26769 for C29H40N2O8Na, Supplementary Figure S4), which is one oxygen atom less than 1 (C29H40N2O9). The 1H and 13C NMR spectra of 2 (Supplementary Figures S5 and S6) were very similar to those of 1.18 Comparison of the NMR data of 2 with those of 1 revealed that the only difference between the two compounds was replacement of the 6-methoxy group in 1 by the 6-methyl group in 2, which was confirmed by the 2D NMR data analysis of 2. In particular, the 1H-1H COSY correlations of H-5/H-6/H-7, H-6/H3-23 and HMBC correlations of H3-23/C-4, C-6, C-5, in combination with the shifts of these proton and carbon resonances established the CH3-6 in 2. Therefore, the structure of 2 was determined to be 6-demethoxy-6-methylgeldanamycin (Figure 1). The NMR chemical shifts of 2 were assigned completely by HSQC, COSY and HMBC spectroscopic data (Supplementary Figures S7–S10) as indicated in Table 1.

Table 1 The NMR spectra data of 2

Compound 2 is a shunt product in 1 biosynthesis. Compound 1’s biosynthesis consists of a starter unit (3-amino-5-hydroxybenzoic acid) assembly, extender units (one malonyl, two 2-methoxymalonyl and four 2-methylmalonyl units for polyketide chain building) condensation and tailoring modifications.17, 18, 19 Obviously, 2 is derived from mis-incorporation of a 2-methylmalonyl unit in place of normal incorporation of a 2-methoxymalonyl unit into the polyketide chain in 1 biosynthesis. It is interesting to note that macbecin (or ansamitocin) as a close 1 analog from Actinosynnema pretiosum ATCC 31280 (or 31565) also contains a methyl (not a methoxyl) side group at C-6 of its polyketide chain, indicating that a 2-methylmalonyl unit incorporates into the polyketide chain in the corresponding condensation reaction of macbecin (or ansamitocin) biosynthesis.20, 21

Compound 2 must share the same absolute configurations with 1, as 2 is co-produced with 1 and must be biosynthesized by the same set of enzymes as 1.22, 23, 24 In particular, the incorporation of 2-methylmalonyl (not 2-methoxymalonyl) unit into the polyketide chain for 2 biosynthesis should not change the configuration of C-6, which is presumably determined by the enoylreductase domain of module 5 of polyketide synthases (PKS) for 1 biosynthesis.22

The identification of 2 in 1 preparation indicates that the acyltransferase domain of module 5 (AT5) of PKS for 1 biosynthesis shows a promiscuous substrate specificity for 2-methoxymalonyl CoA and, to a small degree, 2-methylmalonyl CoA. This phenomenon is observed occasionally in the biosynthesis of some microbial polyketides. For examples, the two components (A and B) of galbonolide are derived from substrate tolerance (for 2-methoxymalonyl and 2-methylmalonyl CoA) of AT5 of PKS in galbonolide biosynthesis.25 The two components (A and B) of epothilone are also derived from relaxed substrate specificity (for malonyl and 2-methylmalonyl CoA) of acyltransferase domain of module 3 of PKS in epothilone biosynthesis.26 Recently, three rapamycin analogs (as impurities) were reported, with each one resulted from mis-incorporation of a 2-ethylmalonyl unit (in place of normal incorporation of a 2-methylmalonyl unit) into the polyketide chain at corresponding positions by the acyltransferase domain of module 3, 7 or 13 of PKS in rapamycin biosynthesis.27

To date, it is still difficult to foretell whether an acyltransferase domain in modular PKS possesses promiscuous substrate specificity or not, or even to predict its substrate specificity for 2-methoxymalonyl CoA, because only a few such acyltransferase domains are reported. We believe that the AT5 of PKS for 1 biosynthesis may be useful in establishing an in silico method to predict substrate promiscuity of acyltransferase domains of modular PKS in the future.

Patel et al.28 reported an 1 analog 6-desmethoxygeldanamycin (KOSN1631, Figure 1) produced by an AT5-engineered strain of S. hygroscopicus, in which the AT5 of 1 PKS was replaced with the acyltransferase domain (with a substrate specificity for malonyl CoA) of module 2 of rapamycin PKS by genetic recombination. Compared with 1, KOSN1631 showed a significant decrease in cytotoxicity against human breast adenocarcinoma cell line SKBr3, with an IC50 value of 3.2 μM (for 1, 0.041 μM). We conducted a preliminary cytotoxicity assay of 2 against human liver hepatocellular carcinoma cell line HepG2 by MTT (3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide).29 Compound 2 also exhibited lower cytotoxicity than 1 against HepG2, with an IC50 value of 10.5 μM (for 1, 0.37 μM), suggesting that the 6-methoxy group may have an important role in 1’s cytotoxicity against cancer cells.