Biosynthesis of Methanofuran in Methanobacterium thermoautotrophicum"

The 13C NMR signals of methanofuran were assigned by two-dimensional 'H and 13C NMR experiments. On this basis, the incorporation of 13C-labeled acetate and pyruvate into methanofuran by growing cells of Methanobacterium thermoautotrophicum was analyzed by one- and two-dimensional 13C NMR experiments. The data were analyzed by a retrobiosynthetic approach based on a comparison of labeling patterns in a variety of metabolites. The data show that the furan ring is formed by condensation of two molecules from the pyruvate/triose pool. The tetracarbocylic acid moiety is assembled from ketoglutarate, two molecules of ace- tyl CoA, and one molecule of carbon dioxide. to hydrolyze RNA. Nucleosides were isolated from the supernatant as described (Eisenreich et al., 1991a). The protein-containing residue was hydrolyzed in 6 M HC1, and amino acids were isolated as described earlier (Eisenreich et al., 1991a, 1991b).

by NMR spectroscopic techniques. This paper reports on a detailed NMR investigation of methanofuran and on the incorporation of 13C-labeled acetate and pyruvate.
[1-"C]Pyruvate was obtained from Isotec Inc. (Miamisburg, OH). Other chemicals were of the highest purity available.
Microorganism-Methanobacterium thermoautotrophicum Marburg (DSM 2133) was obtained from the Deutsche Sammlung von Mikroorganismen, Braunschweig, Federal Republic of Germany. The strain was subcultured at weekly intervals in serum bottles under an atmosphere of H2/COz (4:1, v/v) which were incubated at 65 "C.
Bacterial Culture-M. thermoautotrophicum was grown on mineral salt medium under an atmosphere of HZ/CO2 as described earlier (Eisenreich et al., 1991a). 'T-Labeled acetate and pyruvate were added to logarithmically growing cultures to a final concentration of 4.0 and 2.1 mM, respectively. The pH was continuously adjusted to 6. Cells were harvested at the end of exponential growth phase.
All I3C NMR spectra were recorded under identical conditions. Data acquisition and processing parameters were as follows: 64,000 data set, 30" pulse width (2 ps), 2.5-s scan interval, 14.7-kHz spectral range, composite pulse decoupling, 1-Hz line broadening.
Two-dimensional double quantum filtered COSY, distortionless enhancement by polarization transfer, and INADEQUATE experiments were performed according to standard Bruker software (DISR87). Phase-sensitive two-dimensional TOCSY' and spin-locked NOE (ROESY) spectroscopy were performed according to Braunschweiler and Ernst (1983) and Bothner-By et al. (1984). The 'Hdetected multiple quantum 'H-% chemical shift correlation experiments were performed according to Bax and co-workers (Bax and Subramanian, 1986;Bax and Summers, 1986). Samples were not rotated during two-dimensional experiments. Water suppression was achieved by phase-coherent presaturation.
Data acquisition and processing parameters for two-dimensional experiments were as follows. COSY: 32 scans per tl increment, 1.6-s relaxation delay, 512 X 2056 raw data matrix size zero-filled to 2056 in t, and processed with 3 Hz Gaussian in the f l dimension and 90"shifted sine bell filtering in the fz dimension. TOCSY: 24 scans per tl ' The abbreviations used are: TOCSY, total correlation spectroscopy; HMQC, 'H-detected heteronuclear multiple quantum coherence; HMBC, 'H-detected multiple bond heteronuclear multiple quantum coherence; ROESY, rotating frame nuclear Overhauser and exchange spectroscopy; NOE, nuclear Overhauser effect; HPLC, high performance liquid chromatography. increment, 2.0-s relaxation delay, 62-ms MLEV-17 mixing period preceded and followed by 2.5-ms trim pulses; 90" pulse width, 60 ps; 512 X 2056 raw data matrix size, zero-filled to 2056 in fl and processed with 2 Hz Gaussian in the fl dimension and 90"-shifted sine hell filtering in the fz dimension. ROESY: 64 scans per tl increment, 1.0s relaxation delay, 200-ms continuous wave spin lock period; 90" pulse width, 60 ps; 360 X 1024 raw data matrix size, zero-filled to 1024 in tl and processed with 2 Hz Gaussian in the fl dimension and 60"shifted sine bell filtering in the fz dimension. 'H-l'C HMQC: 64 scans per t, increment, start of coherence experiment 123 ms after bilinear rotation decoupling pulse, 3.5-ms delay period for evolution of 'JCH corresponding to a coupling constant of 145 Hz; I3C decoupling during acquisition by the globally optimized alternating-phase rectangular pulses sequence; 500 X 2048 raw data matrix size, zero-filled to 1024 in tl and processed with 10 Hz Gaussian in fz and 90O-shifted squared sine filtering in f2.  scans per tl increment, 0.8-s relaxation delay, 3.5-ms delay for suppression of 'JcH; 60-ms delay period for evolution of long range couplings; 200 X 1024 raw data matrix size, zero-filled to 512 words in tl and processed with 90"shifted sine bell filtering in fl. INADEQUATE: 128 scans per tl increment, 2.0-s relaxation delay, Ernst-type phase cycle, 3.6-ms delay for evolution of 'Jcc, 128 X 512 raw data matrix size, zero-filled to 512 words in tl and processed with 6O"-shifted sine bell filtering in fl and f2.
High Performance Liquid Chromatography-Methanofuran was analyzed by reversed-phase HPLC using a column of Nucleosil 10 Cle (4.55 X 250 mm) and an eluent containing 13% methanol and 100 mM ammonium formate. The effluent was monitored by photometry (220 nm) and by refractometry. The retention volume of methanofuran was 9.2 rnl.
Isolation of Methanofuran-Cell paste of M. thermoautotrophicum (100 g) was suspended in 50% aqueous acetone (400 ml) at -10 "C. The suspension was stirred at 4 "C for 30 min. The suspension was centrifuged, and the residue was again extracted as described until the supernatant was colorless (approximately 10 times). The supernatants from the extraction procedures were combined and concentrated to dryness under reduced pressure. The residue was dissolved in 80 ml of water and placed on a column of QAE Sephadex A-25 (HCO, form, 3 X 41 cm). The column was developed with a linear gradient of 0-1.5 M NH,HCO, (total volume, 2 liters). Fractions of 20 ml were collected. Methanofuran was eluted at 1.4 M NH4HCO3. Fractions containing methanofuran were adjusted to pH 6, concentrated to dryness under reduced pressure, and lyophilized.
Purification of Methanofuran-Crude methanofuran was purified by preparative HPLC using a column of Lichrosorb RP18 (16 X 250 mm) with an eluent containing 22% methanol and 30 mM formic acid. The effluent was monitored by photometry (220 nm) and by refractometry. Methanofuran had a retention volume of 500 ml. Fractions were combined and evaporated to dryness under reduced pressure. Preparative HPLC purification was repeated using the same column with an eluent containing 33% methanol and 60 mM formic acid. The retention volume was 145 ml. Fractions were combined, evaporated to dryness under reduced pressure, and dried under vacuum.
Isolation of Nucleosides and Amino Acids-The residual cell mass after extraction of methanofuran was treated with 1 M NaOH to hydrolyze RNA. Nucleosides were isolated from the supernatant as described (Eisenreich et al., 1991a). The protein-containing residue was hydrolyzed in 6 M HC1, and amino acids were isolated as described earlier (Eisenreich et al., 1991a(Eisenreich et al., , 1991b.

RESULTS
NMR Spectroscopy-Biosynthetic studies monitored by I3C NMR spectroscopy require unequivocal 13C NMR signal assignments for all carbon atoms of the target molecules. Leigh et al. (1984) published 'H and I3C NMR spectra of methanofuran, but a rigorous NMR analysis of methanofuran has not been reported hitherto. As a prerequisite for the biological studies reported in this paper, a detailed NMR analysis of methanofuran was required.
Several areas of the I3C NMR spectrum of methanofuran are crowded. In an attempt to minimize overlapping of 13C signals, we recorded I3C NMR spectra of methanofuran in different solvents and at different pH values. The most favorable conditions were found in water or D20 at pH 3.6. All subsequent NMR experiments with methanofuran were performed under these conditions. The 'H NMR spectrum of methanofuran was assigned by conventional two-dimensional NMR methodology. Initially, 'H NMR signals of methanofuran were grouped into networks of scalar-coupled protons by a TOCSY experiment with a relatively long mixing period of 67 ms. Examination of the TOCSY spectrum revealed five scalar-coupled 'H spin systems representing the protons of the phenolic ring (3b/5b, 2b/6b), the ethylamine moiety (7b, 8b, NHb), each of the two glutamyl groups (4c, 3c, 2c, and NHc; 4d, 3d, 2d, and NHd), and the tetracarboxylic moiety (2e to 7e) ( Table I). Direct scalar connectivities were then determined by a double quantum filtered COSY experiment ( Table   I). No COSY or TOCSY connectivities were observed for the furan system. Amide protons were detected by comparative measurements in H20 and D'O. The signals at 7.93,8.07, and 8.11 ppm were found to disappear in D20 by exchange with the solvent.
Through space proton interactions were determined by a spin-locked NOE experiment (2D-ROESY). Nuclear Overhauser connectivities via the amide protons afforded connections between the scalar-coupled spin systems. Specifically, the triplet signal at 7.93 ppm, which was identified as NHb by the TOCSY and COSY data, gave positive NOE crosspeaks to the signals at 3.43 and 2.77 ppm (protons 8b and 7b, respectively, from the TOCSY data) and to a signal at 2.27 ppm belonging to a glutamyl moiety (Fig. 2). On this basis, the spin system of one glutamate moiety (2c-5c, NHc) could be identified unequivocally.
The NH signals at 8.07 ppm (NHc) showed NOE connectivities to protons of both glutamyl moieties. Similarly, the amide proton at 8.11 ppm (NHd) gave NOE interactions to the signals of 3d and 2e, respectively ( Fig. 2 and Table I).
Thus, all five scalar-coupled spin systems could be connected.
It should be noted that the signals of 2e and 7e were unequivocally assigned despite the inherent symmetry of the free tetracarboxylic acid. The proton assignments are summarized in Table I. The signals of the protons 4e/5e, 4d/7e, and 3e/6e could not be resolved. Connectivities for the protons of the furan module could not be obtained by 'H NMR experiments.
With the 'H NMR assignments at hand, it was possible to obtain assignments of the I3C NMR signals by 'H-13C correlation experiments. Proton multiplicities of 13C NMR signals were determined by distortionless enhancement by polarization transfer experiments (Table 11). CH connectivities were established by 'H-detected two-dimensional 'H-13C shift correlation experiments for reasons of sensitivity. One-bond CH connectivities were established by a HMQC experiment permitting the assignment of carbon atoms with attached protons ( Fig. 3 and Table 11). Two-bond and three-bond CH connectivities were established by a HMBC experiment (  Referenced to external 3-(trimethylsilyl)-l-propanesulfonic acid, sodium salt, in D20 (pH 3.6).  Table I could be observed by inspection of the two-dimensional matrix at a lower contour level. The one-dimensional 'H NMR spectrum of methanofuran is shown a t the axes.
Additional confirmation of 13C assignments was obtained by an INADEQUATE experiment performed with a sample of methanofuran from a growth experiment with [1,2-"CC,] acetate ( Fig. 4 and Table IT). Carbon pairs biosynthetically derived from acetate gave strong 13C-13C correlation signals. This experiment corroborated the assignments of 10 pairs of directly adjacent carbon atom pairs which had been jointly incorporated from the double-labeled precursor. It should be noted that some of the 13C-13C couplings in Fig. 4 are nonlinear as a consequence of the relatively small chemical shift differences between the coupled carbon atoms. As a consequence, the satellites are not symmetrical with respect to the unlabeled central peak.
The carbon atom pairs lc/ld (179.09 ppm) and 9e/10e (180.97 and 181.00 ppm) could not be assigned on the basis of two-dimensional correlation methods due to signal overlapping. However, a detailed analysis of '%-13C coupling satellites in the one-dimensional 13C NMR spectrum obtained from the fermentation experiment with [ 1,2-'3C2]acetate revealed 'nC-'3C coupling from C-5e to the signal at 181.00 ppm, which was therefore assigned as C-lOe.
Biological Studies-". thermoautotrophicum (Marburg strain) was grown autotrophically in a mineral salt medium at 65 "C. The culture medium was gassed with a mixture of H,/CO2 (80:20, v/v). The growing cultures were supplemented with 13C-labeled acetate (4.0 mM) or pyruvate (2.1 mM). The pH was kept at 6.0 to facilitate the uptake of the organic supplements.
After a culture period of 50-100 h, the cells were harvested by centrifugation. Methanofuran was obtained from the cell mass by repeated extraction with acetonelwater. The raw material was purified by anion-exchange chromatography followed by preparative reversed-phase HPLC.
I3C NMR spectra of methanofuran were recorded a t 90.6 MHz (Fig. 5). The acquisition parameters were the same throughout the study. Integrals of NMR signals were determined. Relative 13C abundances were calculated for all carbon atoms of methanofuran analyzed by comparison of signal integrals with the integrals of a natural abundance sample. The carbon atom with the lowest relative 13C enrichment in each compound was arbitrarily assigned a value of 1.0 under the assumption that at least one carbon atom in each compound was derived exclusively from CO, and not Referenced to external 3-(trimethylsilyl)-l-propanesulfonic acid in DzO (pH 3.6). from acetate or pyruvate. This assumption has been shown t o be valid in studies with M. thermoautotrophicum but need not be correct in studies with other anaerobic microorganisms (Strauss et al., 1992).
The data are summarized in Table  111. The carboxylic groups IC, Id, 9e, and 8e of methanofuran were unlabeled in all experiments. Thus, these carbon atoms were obligatorily derived from COz and could not be contributed by either acetate or pyruvate.
Some of the signals of methanofuran were closely adjacent even at pH 3.6. Due to signal overlap, it was sometimes impossible to obtain separate enrichment values for overlapping signals. This presented no problems in cases where all overlapping signals were in the natural abundance range. For example, neither of the overlapping signals of C-lc and C-ld acquired significant amounts of label from any of the precursors studied. It follows that these carbon atom were derived from carbon dioxide.
The problem of signal overlap was more prominent in the sample obtained from [ 1,2-'3C2]acetate because contiguously labeled carbon pairs appear as multiplets containing a central signal corresponding to the singly labeled species and satellites due to 13C-'3C coupling. However, the signal overlap did not significantly reduce the biosynthetic information for two reasons. (i) The information obtained from [ 1,2-"C2]acetate, on one hand, and from [1-'3C1]-and [2-13Cl]acetate, on the other hand, is redundant to a considerable extent. (ii) The overlapping multiplets in the one-dimensional 13C NMR spectrum of the sample obtained from double-labeled acetate could be resolved by the INADEQUATE experiment shown in Fig.  4. In this double quantum experiment, only pairs of labeled carbon atoms are detected. The f 2 dimension corresponds to the chemical shift of single carbon atoms, but the ti dimension represents the sum of the chemical shifts of two contiguous and l3C-labeled carbon atoms. Thus, even if the signals overlap in the t? dimension, they are clearly separated in the ti dimension. By this experimental approach, each acetate moiety which had been incorporated into methanofuran as an intact unit could be determined unequivocally.
The experiment with [l-13C]pyruvate needs specific consideration, because these data were not redundant with the other experiments. Luckily, several cases of signal overlap concerned pairs of atoms which were both unlabeled (e.g. 9e/10e, lc/ld, 4e/5e). The only uncertainty with biosynthetic relevance concerned the unresolved groups of signals for le, 5c, and 5d in the methanofuran signal from [ l-'"C]pyruvate. This issue will be discussed in detail below.
Two pairs of contiguous carbon atoms were incorporated from [1,2-"C2]acetate into the furan ring, and three pairs were incorporated into the tetracarboxylic moiety. Each of the glutamate moieties showed one pair of carbon atoms from acetate. About 60-80% of the signal intensity of these carbon atoms was located in the satellites arising by I3C-l3C coupling as shown by evaluation of 13C signal integrals. These carbon atom pairs were also visualized by a two-dimensional INADEQUATE experiment (Fig. 4 and Table 11) and by a '"C TOCSY experiment (data not shown). The results of all labeling experiments are summarized in Fig. 6. Joint incorporation of contiguous carbon atom pairs are shown by bold lines.
We have described earlier the labeling patterns of amino acids and nucleosides biosynthesized in M. thermoautotrophicum from 13C-labeled acetate and pyruvate (Eisenreich et al.,  1991a, 1991b). The glutamate moieties of methanofuran virtually duplicate the labeling pattern of glutamate (6, Fig. 7) isolated from cell protein. Similarly, the labeling pattern of the 4-hydroxyphenylethylamine moiety repeats the labeling pattern of tyrosine (Eisenreich et al., 1991a). This correspondence of labeling patterns is biosynthetically trivial but documents the validity of the experimental approach. The labeling patterns of amino acids and nucleosides described earlier allow the reconstruction of central metabolite pools such as pyruvate (2) and dicarboxylic acids by a retrobiosynthetic approach (Fig. 7). For example, the labeling pattern of precursors from the triose/pyruvate pool could be reconstructed from the labeling pattern of ribose (3) and amino acids Bacher, 1991a, 1991b). This analysis served as a basis for the interpretation of the labeling pattern of methanofuran by a pattern-recognition approach.
The labeling pattern of the furan moiety suggests the incorporation of two triose modules. More specifically, a plausible reaction mechanism can be written with dihydroxyacetone phosphate ( 7 ) and pyruvate (2) as precursors (Fig. 8). These findings are in agreement with the results obtained by mass spectrometric studies (White, 1988).
Three contiguous pairs of carbon atoms are incorporated into the tetracarboxylic subunit of methanofuran. Notably, one of the incorporated acetate modules contributes the carboxylic group 10e. This finding is firmly documented by the two-dimensional INADEQUATE and 13C TOCSY experiments.
Both  Fig. 7). In  Stupperich, 1978, 1980). The carbon atoms 6e and 7e are derived from an acetate module, and thus one would expect that the carbon atoms 6e-8e are formed via pyruvate as intermediate. However, this hypothesis can be ruled out since the carboxylic group 8e is not labeled from C-1 of pyruvate. It follows that the carboxylate group 8e is derived obligatorily from carbon dioxide.
The observed labeling pattern suggests the following biosynthetic pathway (Fig. 9). An aldol-type reaction between 2ketoglutarate ( 5 ) and acetyl-coA yields the hydroxy acid 8. Elimination of water yields the unsaturated acid 9 which adds a second molecule of acetyl-coA. Reductive carboxylation via Reconstruction of central metabolites from 13C enrichments in amino acids and nucleosides by a retrobiosynthetic approach (from Bacher, 1991a, 1991b).
The arrows indicate reconstructed precursors in analogy to the retrosynthesis concept in organic chemistry. For details see Fig. 6. the keto acid 10 and the hydroxy acid 11 yields the product 12.
The proposed reaction sequence does not predict whether the biosynthetic pathway proceeds at the level of the free acid or starts from a carboxamide-type precursor. In either case, a nonsymmetrical distribution of label would result as a consequence of the prochiral character of the meso-tetracarboxylic acid.

DISCUSSION
All 13C NMR signals of methanofuran have been unequivocally assigned by a combination of one-and two-dimensional experiments. Notably, unique assignments were possible for the carbon atoms of the tetracarboxylic moiety despite its inherent symmetry. On this basis, the labeling pattern of methanofuran biosynthesized from 13C-labeled acetate and pyruvate was determined by 13C NMR spectroscopy.
Retrobiosynthetic analysis of biosynthetic amino acids and nucleotides gave the labeling patterns of central metabolic intermediates with high accuracy as documented by statistical analysis (Eisenreich and Bacher, 1991b). On this basis, the labeling of the methanofuran molecule was interpreted by a pattern-recognition approach. Additional confirmation of the validity of the experimental method was obtained by the close correspondence of the labeling patterns of the two glutamyl moieties and the 4-hydroxyphenylethylamine moiety in the coenzyme with glutamate and tyrosine from protein, respectively.
Furan ring systems have been observed in a variety of natural products. The furan module of reductiomycin is formed by a complex rearrangement reaction from 4-hydroxybenzoate (Beale et al., 1986). However, 4-hydroxybenzoate is clearly not a precursor of the furan ring of methanofuran. This is immediately obvious from the labeling pattern of methanofuran in the growth experiment with [1,2-l3CZ]acetate. Due to the symmetry of the phenol ring, coupling is observed between all ring atoms of phenol with the exception of the para position. On the other hand, the acetate moieties appear localized in the furan ring of methanofuran.
The labeling pattern of the furan module reflects the incorporation of two molecules from the triose/pyruvate pool. A plausible reaction mechanism starting from dihydroxyacetone phosphate (7) and pyruvate (2) suggests the formation of the ring system by an aldol condensation (Fig. 8). The amino methyl group could then be formed by reductive amination of the carboxylic group.
The labeling pattern of the tetracarboxylic acid indicates its formation from one ketoglutarate, two acetyl-coA, and one COz. This suggests the sequence of reactions shown in Fig. 9. It should be noted that a-ketosuberate is biosynthesized in methanogenic bacteria by a similar condensation mechanism involving ketoglutarate and two acetyl-coA molecules (White, 1989).
It is surprising that the carboxylic group 8e is derived from COz and not from pyruvate. However, reductive carboxylation is a frequent reaction in methanogenic bacteria and has been demonstrated with acetate, pyruvate, and succinate as substrate Stupperich, 1978, 1980;Fuchs, 1984a, 1984b).
White studied the incorporation of "C-labeled acetate into methanofuran by mass spectrometry (White, 1987(White, ,1988. The advantages of this method are obvious. The sensitivity exceeds the NMR experiment by far, and as a consequence, purification of metabolites can be achieved efficiently by online gas chromatographic separation. However, in contrast to 13C NMR analysis, it is not always possible to determine the enrichment of each individual carbon atom. In agreement with our data, White obtained evidence for the formation of the furan ring system from two three-carbon units (White, 1988). On the other hand, he suggested the formation of the tetracarboxylic moiety from 2-ketoglutarate, and two malonate molecules (White, 1987). The incorporation of malonate into the tetracarboxylic acid subunit of methanofuran is ruled out unequivocally by the NMR analysis reported in this paper.