The chemical origin of the characteristic odor of soil was first investigated in 1891 by Berthelot¹², but not until 1965 was the responsible agent, (−)-geosmin (1), first isolated in pure form from the neutral extract of the fermentation broth of Streptomyces griseus and the structure assigned^4,5. In 1981, incorporation of labeled acetate into geosmin using strains of S. antibioticus suggested that this bicyclic C₁₂ metabolite might be a degraded sesquiterpene¹³. Although there have been more 800 papers dealing with the production, detection, and remediation of geosmin and other volatile metabolites in water supplies, aquaculture products, and wine, there were no further reports on the mechanism of microbial geosmin biosynthesis until five years ago¹⁴.

Expression in Escherichia coli of a 2181-bp gene from S. coelicolor A3(2) (SCO6073) gives a 726-amino acid protein with significant similarity in both the N-terminal and C-terminal halves to the well-characterized sesquiterpene synthase, pentalenene synthase¹⁵. The full length recombinant protein catalyzes the Mg²⁺-dependent conversion of farnesyl diphosphate (FPP, 2) to a mixture of germacradienol (3), germacrene D (4), octalin 5, and geosmin (1), without involvement of any cosubstrates or redox cofactors (Scheme 1)^3,15,16. Incubation of FPP (2) with the closely related recombinant germacradienol–geosmin synthase from S. avermitilis (SAV2163, GeoA) yields an essentially identical mixture of 3, 4, 5, and geosmin (1)¹⁷. Deletion of the S. avermitilis geoA gene abolished both germacradienol and geosmin production, which could be restored by reintroduction of a copy of wild-type geoA¹⁷. Independently, researchers at the John Innes Institute have demonstrated that deletion of the S. coelicolor SCO6073 gene abolishes geosmin formation¹⁸.

Generation of germacradienol (3) and germacrene D (4) results from partitioning of a common intermediate, proposed to be carbocation A, the initial product of ionization and cyclization of FPP (2) (Scheme 1)¹⁶. Formation of germacrene D (4) involves a 1,3-hydride shift of H-1si of FPP¹⁶. The alternative formation of germacradienol (3) involves competing loss of the original H-1si proton of FPP (2) and cyclization to a proposed enzyme-bound, trans-fused bicyclic intermediate, isolepidozene (6), a known compound previously isolated from liverworts¹⁹ that would be converted to germacradienol by proton-initiated ring opening and capture of the resulting homoallyl cation by water. Further conversion of germacradienol to geosmin is proposed to involve protonation–cyclization, a novel retro-Prins type fragmentation resulting in loss of the 2-propanol side chain as acetone, and generation of the octalin intermediate 5. Reprotonation of 5 followed by a 1,2-hydride shift and quenching of the bridgehead cation by water will generate geosmin (1)²⁰. The results of incubation reactions carried out in D₂O are also consistent with this mechanistic proposal³.

Increasing either the concentration of germacradienol–geosmin synthase or the time of incubation enhances the relative proportion of geosmin to germacradienol as well as the absolute yield of 1.^3,17 These results are inconsistent with the formation of germacradienol and geosmin by partitioning of a series of exclusively enzyme-bound intermediates. A substantial fraction of the initially generated germacradienol must be released from the enzyme before rebinding to the protein and cyclization–fragmentation to generate geosmin, without distinguishing whether germacradienol and geosmin formation take place at the same or at distinct active sites, nor whether transient release of germacradienol is a mandatory event in geosmin biosynthesis. We now report conclusive evidence that germacradienol–geosmin synthase is a bifunctional enzyme possessing two independent active sites with distinct catalytic functions.

Two strictly conserved motifs found in all sesquiterpene and monoterpene synthases are essential to binding of the Mg²⁺ cofactor: an aspartate-rich sequence DDXX(D/E), that is found at amino acid (aa) residues 80–120 in microbial synthases and at aa 290–310 in plant synthases, and an NSE triad of residues, (N/D)DXX(S/T)XXXE, generally found 140±5 aa downstream of the aspartate-rich motif^21–25. The vast majority of bacterial and fungal terpene synthases are 330–400 aa in length, corresponding to a subunit M_r 35–45 kDa. The 726-aa germacradienol–geosmin synthase of S. coelicolor is unusual in that it is about twice the size of a typical terpene synthase. Notably, both the N- and C-terminal halves of the SCO6073 protein show significant sequence similarity to the well-characterized sesquiterpene synthase, pentalenene synthase (28% identity, 41% similarity over the aa 1–319; 29% identity, 46% similarity over aa 407–726)¹⁵. Both the N-and C-terminal regions of the protein harbor variants of the canonical aspartate rich domain, with a DDHFLE motif in the N-terminal half and an unusual DDYYP motif in the C-terminal half (Fig. 1). Both halves also display typical NSE motifs at NDLFSYQRE and NDVFSYQKE. Curiously, the N-terminal half also appears to harbor an unusual repeat of the upstream NSE motif, NDVLTSRLHQFE, located 38 aa downstream of the first. A 41-kDa truncated N-terminal mutant of germacradienol–geosmin synthase corresponding to aa 1–366 converts FPP to germacradienol, while a C-terminal truncation mutant (aa 383–726) catalyzes only slow solvolysis of FPP, with no detectable formation of cyclic products¹⁵. Deletion of the N-terminal domain of SCO6073 also abolishes geosmin production by S. coelicolor¹⁸.

Based on the discovery that full-length recombinant SCO6073 protein can convert FPP to germacradienol and geosmin³, we have sought to localize the geosmin synthase activity by examining the behavior of both N- and C-terminal truncation mutants. To this end, we generated a 56 aa-longer C-terminal construct using the NcoI-XhoI fragment of plasmid pRW31 encoding full length SCO6073 as template for PCR amplificaton of the region corresponding to aa M327–H726 (Fig. 1) and purified the derived pJJ3 protein (M_r 46,000), carrying a C-terminal His₆-tag, to >95% homogeneity by Ni²⁺-NTA chromatography.

Incubation of the N-terminal truncated mutant (pRW22p)¹⁵ with 100 μM FPP (2) generated a mixture of germacradienol (3) and germacrene D (4) accompanied by small quantities of octalin 5, but no detectable geosmin (1), as determined by capillary gas chromatography–mass spectrometry (GC–MS) (Scheme 2a, Table 1, Supplementary Fig. 1a). Increasing the concentration of the N-terminal domain from 2.4 to 9.0 μM, conditions under which wild-type protein yields a mixture containing up to 15% geosmin, had minimal effect on the relative proportion of the three products. Significantly, addition of 2.4 μM recombinant C-terminal protein (pJJ3p) to the incubation mixture containing 5.4 μM truncated N-terminal protein and 100 μM FPP resulted in formation of geosmin (1), with a concomitant decrease in the proportion of germacradienol (3) (Scheme 2b, Table 1). Using 5.5 μM C-terminal protein increased the proportion of geosmin (1) to 17% while the fraction of germacradienol dropped to 66%, with insignificant variation in the proportion of germacrene D (4) in both cases and a doubling of the relative proportion of octalin 5 (Supplementary Fig. 1b). Control experiments reconfirmed that incubation of FPP (2) with the C-terminal protein alone generated only minor amounts of the acyclic solvolysis products (E)-α-farnesene (7) and nerolidol (8) along with a trace of (Z)-α-bisabolene (9) (Scheme 2c)¹⁵.

These results established either that the C-terminal protein directly catalyzes the formation of geosmin from germacradienol produced by the N-terminal domain or, less likely, that the C-terminal protein in some way modulates the catalytic activity of the N-terminal domain. To distinguish these two possibilities, we incubated an 85:15 mixture of germacradienol (3) and germacrene D (4), containing ~1% octalin 5, obtained from incubation of FPP with N-terminal protein, with 1.6 μM C-terminal protein (pJJ3p) in the presence of 4 mM MgCl₂ for 6 h at 30 °C. GC–MS analysis indicated a 2 % conversion to geosmin, which increased to 6% when the concentration of C-terminal mutant protein was augmented to 3.1 μM (Scheme 2d; Supplementary Fig. 2). Omission of Mg²⁺ from the incubation mixtures completely abolished geosmin formation.

To explore further the role of the C-terminal domain in the formation of geosmin from FPP and germacradienol (3), we used site-directed mutagenesis to target the conserved downstream aspartate-rich DDYYP and the NDVFSYQKE motifs of the full-length germacradienol–geosmin synthase (Supplementary Table 1). When incubated for 11 h with 49 μM FPP, the D455N/D456N double mutant (19.5 μM) generated a 87:11 mixture of germacradienol (3) and germacrene D (4), accompanied by a trace of octalin 5, without any detectable geosmin. The N598L, D599L, S602A, and E606Q mutants gave similar product mixtures, again with complete suppression of geosmin formation. Importantly, geosmin production could be restored by pre-incubation of 16.1 μM of the D455N/D456N double mutant with 43 μM FPP for 5.5 h at 30 °C, followed by addition of 2.7 μM of the C-terminal truncation protein and incubation for a further 5.5 h, yielding a mixture 89:8:1:2 of 3, 4, 5, and geosmin (1).

Site-directed mutagenesis of the presumptive Mg²⁺-binding domains^21–25 in the N-terminal half of the germacradienol–germacrene D synthase resulted in decreases in catalytic efficiency as well as the generation of a variety of abnormal products that were characterized by GC-MS (Table 2). Thus the D86E mutant generated several aberrant products including the acyclic elimination products (E)-α-farnesene (7) and β-farnesene (10) and a monocyclic sesquiterpene tentatively identified by retention index (RI) and mass spectrum as β-elemene (11) (Scheme 3). Interestingly, the L90D mutant, with the canonical DDHFD sequence, was not only severely impaired catalytically, with a greater than 1,000-fold reduction in k_cat/K_m, but produced primarily a 1:6 mixture of β-elemene (11) and γ-elemene (12). Notably both the S272A and T271A mutants produced normal mixtures of 3, 4, 5, and geosmin (1), with only relatively minor reductions in k_cat/K_m, suggesting that the unusual downstream NSE triad found in the N-terminal half of the protein plays no significant role in catalysis. Most interestingly, the S233A mutant, which was severely impaired catalytically (2500-fold reduction in k_cat/K_m), produced, in addition to the normal products 1 (11%), 3 (57%), 4 (18%), 5 (3%), a new sesquiterpene hydrocarbon, identified as isolepidozene (6) (11%, m/z 204, RI 1483), identified by comparison of the mass spectrum and GC retention index with the data for authentic isolepidozene (Supplementary Fig. 3).

The experiments described above demonstrate unambiguously that the SCO6073 germacradienol–geosmin synthase of S. coelicolor is a bifunctional enzyme in which the N-terminal half of the protein converts FPP (2), the universal acyclic sesquiterpene precursor, to an 85:15 mixture of germacradienol (3) and germacrene D (4). The C-terminal half of the protein then rebinds the germacradienol and catalyzes proton-intiated cyclization–fragmentation of 3 to give geosmin (1). Interestingly, the N-terminal domain also produces small amounts of octalin 5, thought to be an intermediate in the conversion of 3 to 1. Both the N- and C-terminal-catalyzed reactions have an absolute dependence on the divalent cation Mg²⁺. Site-directed mutagenesis clearly implicates the Mg²⁺-binding regions of each half of the protein in the conversion of FPP to germacradienol and thence to geosmin. While the role of Mg²⁺ in terpene synthase catalyzed cyclization of allylic diphosphate substrates is well understood²¹, the role of the divalent cation in the electrophilic polyene-type cyclization is unclear, since the germacradienol substrate does not have a pyrophosphate moiety that might interact with the bound Mg²⁺.

Although bifunctional sesquiterpene and monoterpene synthases have not previously been described, fungal ent-kaurene synthases have two distinct active sites that catalyze successive electrophilic cyclization reactions^26,27. The first reaction in the sequence, known as a Type B reaction, is a proton-initiated polyene cyclization of geranylgeranyl diphosphate (GGPP, 13), catalyzed by the N-terminal domain of the protein, to generate the bicyclic intermediate copalyl diphosphate (CPP, 14) (Supplementary Scheme 1). CPP is then rebound by the C-terminal domain of the diterpene synthase which catalyzes an ionization–cyclization–rearrangement to yield the tetracyclic diterpene ent-kaurene (15) (Type A reaction). Interestingly, the order of the Type A and Type B electrophilic cyclization reactions is reversed in these fungal diterpene synthases compared to S. coelicolor germacradienol–geosmin synthase²⁶. A DVDD motif in the N-terminal domain of ent-kaurene synthase is responsible for the proton-initiated polyene cyclization of the GGPP substrate, similar to the aspartate-rich DVDDTA motif of squalene-hopene cyclase, in which the D376 residue (bold) initiates the Type B polyene cyclization, rather than acting as a Mg²⁺-binding domain²⁸. One or both aspartate residues in the DDYYP motif of SCO6073 may therefore play a similar role in the conversion of germacradienol to geosmin.

The genome sequences of a variety of geosmin-producing actinomycetes and cyanobacteria reveal at least ten additional presumptive geosmin synthases with a high level of sequence homology to SCO6073 (45–78% identity, 57–85% similarity over >700 aa) (Supplementary Table 2). The active site DDHFLE, NDLFSYQRE, DDYYP, and NDVFSYQKE motifs are especially highly conserved in all of these presumed geosmin synthases (Supplementary Fig. 4), as is the unusual, duplicated downstream NSE triad in the N-terminal domain that plays no apparent functional role in FPP binding or catalysis.

The detection of free germacradienol, as well as the demonstration that the C-terminal domain of SCO6073 converts germacradienol, produced by the N-terminal domain, to geosmin establishes that the germacradienol intermediate does not channel directly between the two active sites of the bifunctional protein. Instead substrate transfer involves release of the initially generated intermediate 3 into the medium and diffusive rebinding to the active site of the C-terminal domain.

Previous in vitro biochemical and in vivo molecular genetic studies are in agreement that the N-terminal domain of S. coelicolor SCO6073 is essential for the formation of geosmin^15,18. The definitive demonstration that the N-terminal domain converts FPP to germacradienol but not to geosmin, and that the C-terminal domain catalyzes the cyclization–fragmentation of germacradienol to geosmin is at odds, however, with the earlier report that S. coelicolor mutants carrying an in-frame deletion of aa 380–721 of the C-terminal domain of SCO6073 apparently could still produce geosmin¹⁸.

The significant decreases in k_cat and increases in K_m as well as the formation of aberrant products as a consequence of mutations in the universally conserved Mg²⁺-binding domains of terpene synthases is well-precedented^24,25. Frequently the aberrant products result from premature quenching or derailment of the normal reaction cascade of enzyme-bound carbocationic intermediates. The trans-fused bicyclic sesquiterpene isolepidozene (6) derailment product generated by the S233A mutant is in fact a postulated intermediate in the conversion of the germacradienyl cation A to germacradienol 3 (Scheme 1)^3,17. The structure and stereochemistry of 6 are fully compatible with the observed regiochemistry and stereochemistry of labeling by chirally deuterated FPP of both germacradienol (3) and geosmin (1)^3,16. Further experiments to establish the mechanistic details of the conversion of germacradienol to geosmin and the role of the protein in mediating this fascinating transformation are in progress.

Supporting Information