Ethics statement

Nine healthy human volunteers were recruited at the Medical University of Graz. Written informed consent was obtained from all participants, and stool samples were collected and pseudonymized by Dr C. Högenauer. The study was conducted according to federal guidelines, the local ethics committee regulations and the Declaration of Helsinki. This study was approved by the institutional review board of the Medical University of Graz (17–199 ex 05/06).

All mouse studies were performed in accordance with the Commission for Animal Experiments of the Austrian Ministry of Science (GZ BMWFW-66.007/0002-WF/V/3b/2017 and BMWFW-39/12175ex2017/18) and the local ethics committee of the University of Graz. Mice were housed in specific-pathogen-free conditions in individually ventilated cages and maintained on a 12 h light/dark cycle at 21 °C and 48% humidity. Food and water were offered ad libitum. Mice were monitored daily to record their weight and stress levels and killed at the study endpoint with isoflurane and cervical dislocation.

Synthesis of cis/trans-TM

For the current study, the synthetic TV batch was identical to that in ref. 12, but TM was re-synthesized. Three equally sized batches were prepared consecutively as described in the following procedure: A dry 40 ml Schlenk tube with magnetic stirring bar was charged with a solution of (S)-(2-amino-3-hydroxyphenyl)(2-(hydroxymethyl)pyrrolidin-1-yl)-methanone (200 mg, 846 µmol, 1.0 eq.) in 2.8 ml of a 1:1 mixture (v/v) of anhydrous CH2Cl2/DMSO. The brown reaction mixture was cooled to 0 °C (ice bath), 370 µl (2.12 mmol, 2.5 eq.) DIPEA was added, and the orange solution was stirred for 5 min, after which 337 mg (2.12 mmol, 2.5 eq.) SO3·pyridine was added in one portion (brown solution). The reaction mixture was stirred at 0 °C for 60 min. After warming to room temperature, CH2Cl2 was removed under reduced pressure at 23 °C (Schlenk line with preceding cooling trap). The crude brown oil was divided into two high-performance liquid chromatography (HPLC) vials, and the product was isolated via preparative reversed-phase HPLC. The synthesis procedure was repeated two times. The pure fractions of all batches were pooled and lyophilized. The resulting yellow lyophilisate (197 mg) was further purified via flash column chromatography (SiO2, CH2Cl2/MeOH = 20:1 (v/v) to yield the pure title compound as a pale-yellow solid. Yield: 156 mg (666 µmol, 26%), pale-yellow solid C12H14N2O3 (234.26 g mol−1). Rf = 0.38 (CH2Cl2/MeOH = 10:1 (v/v); staining: CAM).

Preparative and analytical HPLC method

RP Macherey-Nagel 125/21 Nucleodur 100-5 C18ec column (21 × 125 mm, 5.0 µm particle size, 100 Å pore size), column oven: 26 °C, flow rate 15 ml min−1, product detection: 220 nm; 0.0–7.0 min MeCN/H2O = 2:98 (v/v), 7.0–82.0 min linear increase to MeCN/H2O = 50:50 (v/v), 82.0–90.0 min linear increase to MeCN/H2O = 95:5 (v/v), 90.0–100.0 min hold MeCN/H2O = 95:5 (v/v), 100.0–102.0 min return to initial conditions. Analytical HPLC method: 40 °C, flow rate 0.7 ml min−1; 0.0–0.5 min MeCN/0.05% TFA = 10:90 (v/v), 0.5–9.0 min linear increase to MeCN/0.05% TFA = 100:0 (v/v), 9.0–11.5 min hold MeCN/0.05% TFA = 100:0 (v/v), 11.5–12.0 min return to initial conditions. HPLC purity (>95%) after prepHPLC and subsequent flash column chromatography is shown in Supplementary Information (Supplementary Table 3 and Supplementary Figs. 1 and 2).

Growth inhibition of faecal bacteria

Fresh human stool (3 g) was suspended in 30 ml phosphate-buffered saline (PBS) and filtered through gauze. Flow-through plus a 10 ml PBS wash was collected and centrifuged at 550g for 15 min. The pellet was resuspended in 3 ml PBS. One 1 ml aliquot was incubated with 170 µM (40 μg ml−1) TM and a second 1 ml aliquot with solvent (EtOH, 0.5%) for 2 h at 37 °C under anaerobic conditions. Conversion of faecal quantities of analytes typical for mouse colonization (nmol TM g−1 or pmol TV g−1) to concentration was approximated by setting 1 g wet weight to 1 ml. Faecal samples (n = 16) of in-house-bred 8-week-old adult female C57Bl/6J mice housed in four cages (n = 4 mice each) were pooled for each cage, then homogenized in 16 ml PBS by vortexing with glass beads. Debris was allowed to sediment briefly before the supernatant was centrifuged at 550g for 15 min. The pellet was resuspended in 3.2 ml PBS, and 1 ml aliquots were incubated with 170 µM TM, 17 µM (6 μg ml−1) TV or EtOH for 2 h at 37 °C under anaerobic conditions. Treated human and mouse samples were washed, resuspended in 1 ml PBS. Serial dilutions were plated on selective and differential agar media (all purchased from Becton, Dickinson and Company, Austria) and incubated for ≤5 days to culture the following groups: bifidobacteria (Bifidobacterium agar (Bifido)); total anaerobic bacteria (Schaedler agar with vitamin K1 and 5% sheep blood (SCH)); anaerobic Gram-positive bacteria (Schaedler colistin–nalidixic acid agar with 5% sheep blood (CNA)); anaerobic Gram-negative bacteria (Schaedler kanamycin–vancomycin agar with 5% sheep blood (KV)); B. fragilis group (Bacteroides bile esculin agar (BBE)); aerobic Enterobacteriaceae (MacConkey agar without salt (MacC)). Viable cells for each sample and bacterial group were compared and expressed as percentage survival after toxin treatment relative to solvent. Selected colonies from all plates were subcultured and frozen at −80 °C. Isolates were identified via matrix-assisted laser desorption/ionization time-of-flight mass spectrometry analysis, VITEK MS (bioMérieux) and PCR amplification and sequencing of the 16S rRNA gene (V1–V4 region). Primers 27F (5′-AGAGTTTGATCCTGGCTCAG-3′) and 806R (5′-GGACTACCAGGGTATCTAAT-3′) were used with an annealing temperature of 57 °C. Sequences were analysed using Blastn (RefSeq_select).

Bacterial strains and growth conditions

Bacterial strains used in this study are described in Supplementary Table 2. K. oxytoca, K. pneumoniae and P. aeruginosa were cultivated in tryptic soy broth (CASO) or on CASO agar and E. coli in LB broth or on LB agar. K. oxytoca strains used in mouse experiments were toxigenic K. oxytoca gcvA::Km (AHC-6), toxin-deficient K. oxytoca npsB::Tn5-aphA (ΔnpsB), and K. oxytoca npsB::Tn5-aphA with an npsB expression vector (pNpsB)33. Ampicillin-resistant E. coli strain B13 was isolated from faeces of a mock-treated mouse. K. pneumoniae strain J10 and P. aeruginosa J17 were isolated from stool of one human donor. E. coli B13 and K. pneumoniae J10 were plated on CASO-Sm agar to generate the streptomycin-resistant derivatives B13Sm and J10Sm utilized in mouse experiments. Antibiotics (Sigma-Aldrich) were added to final concentrations of 100 μg ml−1 ampicillin (Amp), 40 μg ml−1 kanamycin (Km), 10 μg ml−1 chloramphenicol (Cm), 25 (E. coli) or 50 (K. pneumoniae) µg ml−1 streptomycin (Sm), and 100 (E. coli) or 200 (K. pneumoniae and P. aeruginosa) μg ml−1 rifampicin (Rif). All bacteria were grown at 37 °C.

Monocolonization model and histology

Adult female C57BL/6NRj mice with SOPF status were purchased from Janvier Labs and infected with K. oxytoca at 8 weeks of age. In the first experiment, a total of 16 mice were used (Fig. 2 and Extended Data Fig. 2b–g). Sixteen mice were analysed histologically for Extended Data Fig. 2a. To analyse the longitudinal effects of TM on community structure in detail, we utilized 18 mice (Fig. 3 and Extended Data Figs. 3 and 4). Curam (1,000 mg amoxicillin/200 mg clavulanic acid, Sandoz) was administered in the drinking water (0.4 g l−1) 24 h before infection and until 13 days post-infection. Antibiotic solution was refreshed on day 6 post-infection. To prepare the inoculum, K. oxytoca AHC-6 and ΔnpsB were grown for 24 h on CASO-Km agar and K. oxytoca AHC-6 + pNpsB on CASO-Km/Cm agar. A suspension of each strain was prepared by pooling single colonies in CASO broth to an optical density of 0.1 at 600 nm (OD600). Each mouse was gavaged with 107 cells in 100 μl, or with 100 μl sterile CASO broth (mock). Viable bacteria in inocula were enumerated by plating on selection agar. Mice were kept on antibiotics for 2 weeks and killed after a 7 day antibiotic-recovery phase at day 21. Faecal pellets were collected daily for each mouse to analyse density of K. oxytoca, quantities of TM and TV, and microbiota composition. The absence of intrinsic K. oxytoca in all mouse cohorts was verified by stool culture before infection and randomly for some mice post-infection on Simmons citrate agar with 1% inositol. For histopathological evaluation, colon tissue was excised at sacrifice and epithelial damage was scored as described previously22,63.

Dual colonization models

Curam was administered in the drinking water to in-house-bred 8-week-old adult female C57Bl/6J mice 24 h before infection. For the E. coli/K. oxytoca competition model, 21 mice were gavaged with a mixtures (107:107 cells in 100 µl CASO broth) of strains B13Sm and AHC-6 (n = 11) or B13Sm and ΔnpsB (n = 10). To prepare the inocula, bacteria were grown on CASO agar for 24 h. Separate suspensions of each species were prepared in CASO broth to OD600 = 0.2 and mixed 1:1. Serial dilutions of gavage suspensions were plated on selective agar to determine CFU. Faecal densities of K. oxytoca were determined on CASO-Km plates, and E. coli B13Sm was quantified on CASO containing Sm and 40 μg ml−1 5-bromo-4-chloro-3-indolyl-β-d-galactopyranoside. RifRE. coli were selected on CASO-Sm/Rif plates.

For dual colonization of mice with K. oxytoca and K. pneumoniae, eight in-house-bred 8-week-old adult female C57Bl/6J mice were housed in pairs. Curam was administered in the drinking water to 24 h before infection. Four mice were gavaged with mixtures of strains J10Sm and AHC-6, and four mice were gavaged with J10Sm and ΔnpsB (105:107 CFU in CASO broth). Six days post-infection, antibiotic treatment was switched from Curam to rifampicin (0.1 g l−1 drinking water) after which mice were housed individually. K. oxytoca present in faeces or caecal content were enumerated on CASO-Km, and K. pneumoniae J10 densities were determined on CASO-Sm plates. RifR J10Sm were quantified on CASO-Sm/Rif plates.

Faecal metabolites

The synthesis of 15N-labelled TM and TV used for analyte quantification was described previously32. TM and TV were extracted from faeces and quantified by HPLC high-resolution electrospray ionization mass spectrometry as previously described32. In brief, faecal pellets were collected daily from each mouse and stored in HPLC glass vials at −80 °C. For extraction, pre-weighed samples were homogenized by vortexing. EtOH-dissolved 15N-labelled TM (20 µM) and TV (0.2 µM) were added as internal standards to each sample and vortexed again for 5 min. Samples were evaporated to dryness (10 mbar, 40 °C, 60 min), rehydrated in 20 µl water to maximize signal-to-noise ratio and extracted with 200 µl n-butanol by vortexing for 5 min. Extracts were centrifuged in glass vials (15 min, 4,000g, 20 °C), the organic phase was filtered (0.2 µm, Nylon) and stored at −20 °C until measurement. For each sample set, the lowest measurable TM concentration was used to define the limit of quantification.

Bacterial 16S rRNA gene sequencing

For the data shown in Fig. 2 and Extended Data Fig. 2, DNA of 64 murine faecal samples and 16 caecal samples was isolated with the PowerSoil DNA Isolation Kit (Mo Bio Laboratories). These DNA and two control samples (DNA extraction without faeces) were sequenced in two independent Illumina MiSeq, v3, 2x300bp (2 Mio. read package) runs (run 1, caecal; run 2, faecal) by Microsynth AG. For the longitudinal experiment (Fig. 3 and Extended Data Fig. 3), mouse faecal pellets (n = 108) and two empty tubes (blank) were sent to Microsynth AG for DNA extraction and Illumina MiSeq, v3, 2x300bp (15 Mio. read package) sequencing (run 3; Fig. 3). For library preparation, dual-indexed universal primers 341F (5′-CCTACGGGNGGCWGCAG-3′) and 802R (5′-GACTACHVGGGTATCTAATCC-3′) were utilized in the two-step PCR amplification of the V3–V4 hypervariable regions of the 16S rRNA gene.

Data processing, filtering and statistical analysis

Data received from Microsynth AG was demultiplexed and trimmed for Illumina adaptor residuals and locus-specific primers. In total 6,404,412 (run 1), 4,285,688 (run 2) and 15,413,804 (run 3) reads were generated. Data analysis was performed with QIIME2 (2021.4) (ref. 64) in Miniconda (4.10.3) environment. To obtain amplicon sequence variants, reads were processed with the Divisive Amplicon Denoising Algorithm (DADA2) (ref. 65) with quality score threshold set at 25. Alignments were generated with the q2-dada2 plugin. For sequences obtained in run 1 and run 2, the truncation length was set to 276 for forward and 201 for reverse reads. For run 3, truncation was set to 267 for forward and 181 for reverse reads. Bacterial taxonomy was assigned using RESCRIPt66 and a sklearn-based naïve Bayes classifier67 trained on the SILVA v138 99% 16S full-length database. The first step filtered out features classified as mitochondria and chloroplasts and features detected in empty/blank samples (n = 10 features for run 1 and run 2; n = 3 for run 3). Next, samples with reads below 10,000 were excluded for run 1 and run 2, and samples with reads below 23,000 were excluded for run 3 from further analysis. After this second filtering step, 16 of 16 mouse samples (caecal content, run 1), 53 of 60 faecal samples (run 2) and 90 of 108 faecal samples (run 3) remained for analysis. Amplicon sequence variants were used to generate the tree for the phylogenetic diversity analysis with FastTree2 and MAFFT68,69. Calculation of alpha-diversity indices (observed features, Shannon and Faith’s phylogenetic diversity), beta-diversity indices (Jaccard distance and unweighted UniFrac) and PCoA utilized the q2-diversity pipeline. To standardize the library size across samples, we normalized pre-processed sequencing data by rarefying to a minimum of 10,000 (runs 1 and 2) and 23,000 (run 3) reads per sample, respectively. Recovery rate of unweighted UniFrac was tracked with q2-longitudinal70 with the static point set to day −1 (pre-antibiotic). Statistical test for alpha-diversity indices utilized Kruskal–Wallis test and for beta-diversity calculation pairwise permutational multivariate analysis of variance (PERMANOVA) was used. The P-value correction was done using the Benjamin–Hochberg false discovery rate (FDR) correction. Differences in abundance of bacterial taxa was calculated using LEfSe71. For visualization, GraphPad Prism (9.2.0) and in-house R scripts (version 3.6.2) were used.

Sequence analysis of rpoB

Amplification of rpoB was performed with Phusion-polymerase (New England Biolabs) and genomic DNA (Qiagen DNAeasy Blood & Tissue Kit) or whole cells. Primers rpoB_R4_f (5′-GGATATGATCAACGCCAAGC-3′) and rpoB_4R_r (5′-TCGATAGCAGACAGGTAGTG-3′) generated a 360 bp product (nucleotides (nt) 1,470–1,829) for E. coli and K. pneumoniae. Sanger sequences were obtained with the same primers. Forward and reverse reads were assembled for each clone, and a 333 bp (nt 1,484–1,816) sequence spanning gene cluster I (nt 1,521–1,611) and cluster II (nt 1,686–1,725) was analysed for mutations.

Expression constructions

All plasmids and oligonucleotides used for cloning are listed in Supplementary Table 2. K. oxytoca uvrX and E. coli MG1655 uvrA and mutS were amplified from genomic DNA with the respective primer pairs 1 + 2, 3 + 4, and 5 + 6 (Supplementary Table 2). Amplified genes were placed under control of the araBAD operon promoter in pBAD33, and expression was induced by addition of 0.05% arabinose (Ara).

TM-induced SOS and mutation in vitro

DNA-damage response was monitored in E. coli MG1655 carrying the reporter plasmids pUA66-PrecA-GFP, pUA66-PsodA-GFP, pUA66-PkatG-GFP and pUA66-PsoxR-GFP described previously72. Overnight cultures (ONC) were diluted to OD600 = 0.05 in a 24-well plate containing 1 ml LB medium with varying concentrations of TM. Controls included 28.8 nM ciprofloxacin, 0.25 mM paraquat (Sigma-Aldrich) or 0.5% EtOH. OD600 and fluorescence (GFP 480/510) of these cultures were recorded in a microplate reader (TECAN GENios Pro v3.40 01/06) for 16 h at 37 °C. E. coli MG1655 pUA66 served as autofluorescence control.

For mutagenesis, ONCs of rifampicin-sensitive E. coli MG1655 were diluted to OD600 = 0.05 in LB medium. Aliquots of 2 ml were incubated (16 h, 180 rpm, 37 °C) in glass eprouvettes with TM or EtOH. Cultures were centrifuged for 5 min at 2,300g, and pellets were resuspended in fresh LB medium and incubated for 1 h before plating. To determine the frequency of resistance emergence, undiluted suspensions were plated on LB-Rif agar and serial dilutions were plated on LB agar. Mutation frequencies were expressed as rifampicin-resistant mutants (RifR CFU ml−1) divided by total CFU ml−1. The assay was performed five times independently. Mutations in the rpoB gene were determined as described above for three colonies per assay (n = 15) for each treatment.

E. coli viability assays with Keio collection strains

To assess the impact of TM on viability of E. coli JW0001-1 and JW4019-1, bacterial suspensions (5 × 107 CFU ml−1) were incubated with 170 µM, 85 µM, 42.5 µM TM or 0.5% EtOH in LB for 2 h at 37 °C under aerobic conditions. Serial dilutions for each treatment were plated to determine CFU. Strains carrying the expression plasmids pUvrAEc, pUvrXKo and the vector control were grown overnight in the presence of Cm and Ara. ONCs were diluted with Cm/Ara medium to 5 × 107 CFU ml−1 and incubated in glass eprouvettes with 85 µM TM or with EtOH for 2 h under aerobic conditions. Serial dilutions were plated to determine CFU. Bacterial viability is expressed as percentage of surviving cells in treatment group compared with the solvent control.

Complementation of TM-induced mutation in vitro

E. coli MG1655 harbouring plasmids PBAD-uvrA (pUvrAEc), PBAD-uvrX (pUvrXKo), PBAD-mutS (pMutSEc) or empty vector were grown in LB medium supplemented with Cm and Ara for 5 h at 37 °C × 180 rpm, then diluted to OD600 = 0.05 in LB containing Cm and Ara. Equivalent aliquots were exposed to 42.5 µM TM or EtOH for 16 h at 37 °C × 180 rpm. Serial dilutions were plated with and without rifampicin selection to determine mutation frequencies.

LC50 and mutagenesis assays with human and mouse bacterial isolates

E. coli B13, K. pneumoniae J10 and P. aeruginosa J17 were grown in CASO medium overnight, diluted 1:10 in 5 ml and incubated for 30 min at 37 °C × 180 rpm. Suspensions (OD600 = 0.1) were prepared in PBS, then incubated with varying concentrations of TM for 2 h at 37 °C under anaerobic conditions. EtOH was used as solvent control. Viable cells were quantified with one portion of treated cultures by plating serial dilutions on CASO agar. CASO broth was added to the remaining culture to a final volume of 4 ml, and cultures were incubated for 6 h (E. coli and K. pneumoniae) or 22 h (P. aeruginosa) at 37 °C × 180 rpm. The frequencies of RifR emergence were determined by plating aliquots on CASO agar with and without selection. The mutation spectrum obtained in vitro was determined by sequencing the rpoB alleles of RifR colonies as described above.

Statistical analysis

Collection of human samples was randomized and blinded. No other data collection and analysis was performed blind to the conditions of the experiments. No statistical methods were used to pre-determine sample sizes, but our sample sizes are similar to those reported in previous publications20,72,73. Mice were randomly assigned to respective treatment groups. No animals or data points were excluded from the analyses. Data collection, statistical analysis and visualization utilized Microsoft Excel (22.03) GraphPad Prism (9.2.0), CorelDRAW 2019 (21.0.0.593) and in-house R scripts (3.6.2). The following packages were used with R (3.6.2): ggplot2 (2.3.5), qiime2R (0.99.6), dplyr (1.0.8), tidyverse (1.3.1) and RStudio (1.2.5033). Shapiro–Wilk normality test was applied to all datasets n > 6; for all other sets, we assumed non-normal distribution, but this was not formally tested. Testing for statistical significance always employed two-tailed tests if not stated otherwise.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.