Molecular Basis for Asynchronous Chain Elongation During Rifamycin Antibiotic Biosynthesis

Abstract

The rifamycin synthetase (RIFS) from the bacterium Amycolatopsis mediterranei is a large (3.5 MDa) multienzyme system that catalyzes over 40 chemical reactions to generate a complex precursor to the antibiotic rifamycin B. It is considered a hybrid enzymatic assembly line and consists of an N-terminal nonribosomal peptide synthetase loading module followed by a decamodular polyketide synthase (PKS). While the biosynthetic functions are known for each enzymatic domain of RIFS, structural and biochemical analyses of this system from purified components are relatively scarce. Here, we examine the biosynthetic mechanism of RIFS through complementary crosslinking, kinetic, and structural analyses of its first PKS module (M1). Thiol-selective crosslinking of M1 provided a plausible molecular basis for previously observed conformational asymmetry with respect to ketosynthase (KS)-substrate carrier protein (CP) interactions during polyketide chain elongation. Our data suggest that C-terminal dimeric interfaces-which are ubiquitous in bacterial PKSs-force their adjacent CP domains to co-migrate between two equivalent KS active site chambers. Cryogenic electron microscopy analysis of M1 further supported this observation while uncovering its unique architecture. Single-turnover kinetic analysis of M1 indicated that although removal of C-terminal dimeric interfaces supported 2-fold greater KS-CP interactions, it did not increase the partial product occupancy of the homodimeric protein. Our findings cast light on molecular details of natural antibiotic biosynthesis that will aid in the design of artificial megasynth(et)ases with untold product structures and bioactivities.

Keywords: Biochemistry; Biological Sciences; Cryogenic Electron Microscopy; Enzyme Mechanisms; Polyketide Biosynthesis.

Figure 1.

(A) Biosynthesis of rifamycin precursor, proansamycin X (PAX), by the RIFS assembly line in partnership with RifF and Rif-Orf19 (Mal = malonyl; MeMal = (2S)-methylmalonyl). The clinically-approved antibiotic rifampicin (a.k.a. rifampin) can be chemically synthesized from PAX-derived rifamycin B. (B) The genetic organization of RIFS (rifA–E), rifF, and rif-orf19 in Amycolatopsis mediterranei. (C) Biosynthesis of diketide thioester (d1) en route to PAX by the LM and M1 of RIFS. The absolute stereochemistry of d1 was inferred based on KR sequence motifs that predict stereochemical outcomes of β-ketoreduction and α-epimerization(48). Squiggly lines attached to sulfur atoms depict 4′-phosphopantetheine (Ppant) groups. The third transformation illustrates inter-molecular Claisen condensation between KS- and CP-tethered reactants from different subunits (distinguished by heavy and light shading).

Figure 2.

(A) DBA addition to a homodimeric holo-form PKS module (step 1) followed by quenching with excess thiol (step 2) permits proximity-dependent crosslinking between thiols present on KS and CP domains. For simplicity, only one possible crosslinking outcome is shown, and domains not involved in crosslinking have been omitted (Ppant = 4′-phosphopantetheine). (B) SDS-PAGE analysis following addition of DBA to M1-DD and M1-TEII, which are derivatives of the first PKS module of RIFS, revealed the appearance of low-mobility crosslinked bands (b1–b3) relative to un-crosslinked material (b0). M1-DD contains a C-terminal dimeric docking domain (DD), whereas M1-TEII contains a C-terminal monomeric type II thioesterase (TEII) ‘RifR’ associated with the rifamycin biosynthetic gene cluster(34). (C) Cartoon structures of b0–b3 illustrate their intra- or inter-molecular crosslinked or un-crosslinked nature.

Figure 3.

(A) The overall cryo-EM map of RIFS M1-DD bound to F_ab 1B2 (M1-DD-1B2) in the transacylation-mode and (B) modeled interactions between its 4′-phosphopantetheine (Ppant)-attached CP and AT domains at 3.96 Å gold-standard Fourier shell correlation (GSFSC) resolution (Figs. S5 and S6). Focused refinement produced a supplemental cryo-EM map with more prominent density for the CP domain in the transacylation-mode structure (Fig. S7). (C) The overall cryo-EM map of M1-DD-1B2 in the elongation-mode and (D) modeled interactions between its Ppant-attached CP and KS domains at 3.22 Å GSFSC resolution (Figs. S5 and S8). Side chain coordinates shown in panels B and D were approximated by AlphaFold 3, as the local resolution was insufficient for experimental map-based modeling (Figs. S6 and S8)(41). (E) Comparison of the transacylation- and elongation-mode structures of M1-DD-1B2 highlights potential motion during the catalytic cycle involving a 30° (or 150°) twist of the DH° dimer (or KS-AT dimer) about the pseudo-C2 axis of module symmetry.

Figure 4.

(A) The overall cryo-EM map of crosslinked RIFS M1-TEII bound to F_ab 1B2 (CL-M1-TEII-1B2) at 3.96 Å GSFSC resolution (Figs. S10–S11). A notable region of density corresponding to the KR-CP linker is highlighted in purple. Focused refinement produced a supplemental cryo-EM map with slightly improved resolution of the KR-CP linker (Fig. S12). The equivalent linker in the opposite subunit was not well resolved in either map. (B) Two Arg residues of the KR-CP linker (R1480 and R1481) were predicted by AlphaFold 3 to make electrostatic contacts with D938 and D1078 of the DH° domain from the opposite subunit. (C) WebLogos displaying residue type and frequency at regions containing these residues (i.e., KR-CP linker and DH motifs 1 & 2) were generated from a multiple sequence alignment of 250 homologs of the DH°-KR-CP fragment of RIFS M1 (WP_013222547.1)(49). From this analysis, three out of four of the residues putatively involved in electrostatics are invariant, whereas one of them (R1481) is often replaced by a Pro (see Fig. S13 for a complementary ConSurf analysis of sequence conservation mapped onto the DH°-KR-CP structure)(50).

Figure 5.

(A) Enzymatic production of covalently anchored diketide (d2) by the RIFS LM-M1-X bimodule, where X implies a C-terminal DD or TEII*. KOH and heat treatment of the enzyme-bound product liberated d2′ which was detected by LC-MS/MS. The primary fragment ion of d2′ (m/z = 73, presumed to be propionate; boxed in gray) was harnessed for quantification along with a secondary fragment ion (m/z 117, Fig. S14). (B) Normalized reaction extents are plotted for LM-M1-DD, LM-M1-TEII*, and LM-M1-DD(C802A) based on fragment ion counts over time. The data were fit to a single-phase exponential growth function to obtain approximate rate constants (k_obs). LC-MS/MS quantification of d2′ standards in the presence of LM-M1-DD or LM-M1-TEII* was applied to infer the maximum percent occupancy of each protein with d2 in enzymatic reactions (Fig. S14).