Rifamycin congeners kanglemycins are active against rifampicin-resistant bacteria via a distinct mechanism

Abstract

Rifamycin antibiotics (Rifs) target bacterial RNA polymerases (RNAPs) and are widely used to treat infections including tuberculosis. The utility of these compounds is threatened by the increasing incidence of resistance (Rif^R). As resistance mechanisms found in clinical settings may also occur in natural environments, here we postulated that bacteria could have evolved to produce rifamycin congeners active against clinically relevant resistance phenotypes. We survey soil metagenomes and identify a tailoring enzyme-rich family of gene clusters encoding biosynthesis of rifamycin congeners (kanglemycins, Kangs) with potent in vivo and in vitro activity against the most common clinically relevant Rif^R mutations. Our structural and mechanistic analyses reveal the basis for Kang inhibition of Rif^R RNAP. Unlike Rifs, Kangs function through a mechanism that includes interfering with 5'-initiating substrate binding. Our results suggest that examining soil microbiomes for new analogues of clinically used antibiotics may uncover metabolites capable of circumventing clinically important resistance mechanisms.

Fig. 1

The rifamycin biosynthetic gene cluster and the role of AHBA synthase. a The rifamycin gene cluster from Amycolatopsis mediterranei. b The reaction catalyzed by AHBA synthase and the structure of rifamycin SV (the product of the rifamycin gene cluster). The rifamycin SV structure is colored according to the genes responsible for producing its PK core (red), AHBA-derived substructure (green), and tailoring functionalities (black). The phylogenetic divergence of AHBA synthase genes from previously characterized gene clusters correlates with the different structural classes of ansamycins

Fig. 2

Sequence-based screen for rifamycin congener gene clusters. a Screening overview. DNA isolated from ~1500 soils was screened for the presence of AHBA synthase genes by PCR using degenerate primers. Sequence tags generated in this screen were used to construct a phylogenetic tree, onto which AHBA synthase reference sequences from known rifamycin congener gene clusters were mapped (marked with asterisks). Large, distinct clades in the phylogenetic tree are shown in different colors. Metagenomic DNA cosmid libraries were generated from soils that contained AHBA sequence tags that spanned all AHBA clades predicted to be associated with rifamycin congener gene clusters. To facilitate the recovery of individual clones containing gene clusters of interest, each metagenomic library was expanded to contain >20,000,000 unique eDNA cosmids and formatted as smaller subpools of between 20,000 and 60,000 unique cosmid clones per sub-pool. Primary clones (those containing an AHBA synthase gene) were recovered from AHBA positive subpools using a PCR dilution method and degenerate AHBA synthase primers. The same approach, but with degenerate primers targeting PKS ketosynthase (KS) domains and the rif15A/15B tailoring genes, was used to recover regions of the pathways that flank those found on the primary clone. AHBA sequence tags corresponding with primary clones that were targeted for recovery are indicated with arrows on the phylogenetic tree. b Summary of rifamycin congener gene clusters recovered from the soil metagenomes. Portions of the gene clusters found on primary clones are shown on a gray background

Fig. 3

Analysis of the kng gene cluster and the activity of Kangs A, V1, and V2. a Comparison of the rifamycin (rif) and Kang (kng) gene clusters. Lines connecting the two clusters indicate genes that are predicted to be functionally equivalent. For simplicity, only genes lacking a counterpart in the rif cluster are labeled in the kng cluster. Colored boxes surrounding these genes correspond to the substructures they are predicted to encode (shown in panel C). b Structures of Kangs A, V1 and V2. c Summary of the proposed biosynthesis of Kang V2. The structure of Kang V2 is colored as follows: red, PK core; blue, Emal modification; green, AHBA-derived substructure; black, tailoring modifications. Colored bubbles highlighting the key structural features of Kang V2 correspond with the genes in (A) that are predicted to encode for these features. The PKS module 8 dehydratase (dh) domain, which is predicted to be inactive, is shown in lower case letters to differentiate it from the remaining, active domains. d In vivo activity profiles of the Kangs against Rif^R Sau. The structure of Rif is shown along with the three most commonly mutated RNAP residues in Rif^R Mtb clinical isolates^, . The dashed line and arcs indicate an H-bond and nonpolar contacts, respectively. e. In vitro transcription assay with radiolabeled nucleotides showing the activity of Rif and the Kangs at the concentrations indicated against Msm wild-type and Rif^R βS447L RNAP. F, full-length transcript; A, abortive transcript

Fig. 4

Structural basis for Kang A inhibition of Rif^R RNAP. a Overall view of the Msm TIC bound to Kang A. The Rif scaffold of Kang A is colored orange, K-sugar yellow, K-acid violet. wt, wild-type. The boxed region is magnified in (B) (showing Rif/wt-RNAP), (C) (Kang A/wt-RNAP), and (D) (Kang A/S447L-RNAP), but the view is rotated 90° as shown. b View into the Rif binding pocket (wt-RNAP) from inside the Msm RNAP active site cleft. Carbon atoms of the Rif scaffold are colored orange, carbon atoms of the 1-methyl-piperazine moiety are colored green. The RNAP is shown as a backbone worm with a transparent molecular surface (β, light cyan; β′, light pink). Density for the RNAP active site Mg²⁺ was very weak so it was not modeled, but its position is denoted by a dashed yellow circle. RNAP β subunit side chains that make direct contacts with Rif are shown, labeled and colored cyan. Polar interactions are indicated by dashed lines (gray, H-bonds; red, cation-π interactions). Strong electron density near the positively charged 1-methyl-piperazine group (Supplementary Figure S26) was interpreted as a SO₄⁻ ion. Boxed residue labels denote residues that have been identified as conferring Rif^R when substituted, with red boxes denoting three residues (Msm RNAP β D432, H442, and S447, corresponding to Mtb RNAP β D441, H451, and S456) that comprise the majority of Rif^R substitutions identified from clinical isolates from tuberculosis patients. c. View into the Kang A binding pocket (wt-RNAP). Kang A is colored as in (A). The RNAP is shown as in (A) except RNAP side chains that interact with K-sugar but not Rif (R164, T424) are colored yellow, and R604, which makes a salt bridge with K-acid is colored violet. Residues that confer Rif^R when substituted are denoted by colored boxes as in (B). d View into the Kang A binding pocket (S447L-RNAP). Kang A and RNAP are shown as in (C). The RNAP β subunit segment from 447–450 is distorted, and the loop from 451 to 465 is disordered, resulting in the loss of Kang A/RNAP contacts with β residues L427, L447, L449, G450, and R456

Fig. 5

Kang A contacts with wild-type and Rif^R RNAP. Schematic summary of the Kang A/RNAP β subunit contacts. Residues that make only nonpolar contacts are shown as labels with arcs denoting the contacts. The side chains (or main chain atoms for F430) of residues that make polar contacts are shown in stick format (H-bonds, gray dashed lines; cation-π interactions, red dashed lines). The color-coding of residues/residue labels is as follows: residues that contact the Rif scaffold in the Rif/RNAP structure, cyan; residues that also make nonpolar contacts with K-sugar, yellow arc; residues that contact K-sugar but do not contact the Rif scaffold, yellow. R604 (colored violet) makes a cation-π interaction with the Rif scaffold but also makes a salt bridge with K-acid. Residues that confer Rif^R when substituted are denoted by colored boxes as in (A). Residues that lose contacts with Kang A in the Rif^R S447L RNAP mutant are denoted by red background shading

Fig. 6

Structural basis for Kang A inhibition of iNTP binding. a View of the RNAP active site from the T. thermophilus de novo initiation complex (4Q4Z) with bound Rif superimposed. Shown is the t-strand DNA from +1 to −5 (dark gray), the initiating NTP substrates (i site NTP, ATP; i + 1 NTP, CMPCPP; blue carbon atoms) and two Mg²⁺-ions (yellow spheres; Mg²⁺I is the Mg²⁺-ion chelated in the RNAP active site, Mg²⁺II is bound to the i + 1 NTP). Rif is color-coded as in Fig. 4b. Rif and the NTPs are also shown with transparent van der Waals surfaces. The blue “+” denotes the positive charge of the Rif piperazine moiety, while the red “−” denotes the negative charge of the iNTP γ-phosphate. b Same as (A) but showing Kang A (colored as in Fig. 4c). The negative charge of K-acid is brought in close proximity to the negative charge of the iNTP γ-phosphate. c Sequence of AP3-GU promoter template used in in vitro abortive initiation assays monitoring the effect of Kang A or Rif on RNA dinucleotide synthesis with GTP, GDP, or GMP as the 5′-initiating nucleotide. The initial transcribed sequence of the Mtb AP3 promoter (top) was engineered to allow only RNA dinucleotide synthesis (5′-GU-3’) in the presence of GTP, GDP, or GMP as the 5′-initiating nucleotide and UTP. The mutated bases are denoted in bold italic (AP3-GU, bottom). d Kang A or Rif inhibition of in vitro abortive initiation of RNA dinucleotide synthesis using the AP3-GU promoter template; (top) 1 mM GTP + 50 μM α-³²P-UTP; (middle) 2 mM GDP + 50 μM α-³²P-UTP; (bottom) 4 mM GMP + 50 μM α-³²P-UTP. e Plotted is the RNA dinucleotide synthesis with Kang A relative to the same condition with Rif, normalized by the results with no antibiotic. Kang A has a strong inhibitory effect with GTP as the 5′-initiating nucleotide (blue bars), a weaker effect with GDP (red bars), and no inhibitory effect with GMP (green bars). The error bars denote standard error of four measurements

Fig. 7

Model for the evolution of a structurally complex Kang family molecule. A stepwise increase in the structural complexity of the antibiotic is envisioned to result from a series of horizontal gene transfer events. Genes acquired at each step are shown in boxes and are highlighted according to the structural feature they are predicted to encode