Abstract
Antibiotics are used for fighting pathogens, but also target our commensal bacteria as a side effect, disturbing the gut microbiota composition and causing dysbiosis and disease1-3. Despite this well-known collateral damage, the activity spectrum of the different antibiotic classes on gut bacteria remains poorly characterized. Having monitored the activities of >1,000 marketed drugs on 38 representative species of the healthy human gut microbiome4, we here characterize further the 144 antibiotics therein, representing all major classes. We determined >800 Minimal Inhibitory Concentrations (MICs) and extended the antibiotic profiling to 10 additional species to validate these results and link to available data on antibiotic breakpoints for gut microbes. Antibiotic classes exhibited distinct inhibition spectra, including generation-dependent effects by quinolones and phylogeny-independence by β-lactams. Macrolides and tetracyclines, two prototypic classes of bacteriostatic protein synthesis inhibitors, inhibited almost all commensals tested. We established that both kill different subsets of prevalent commensal bacteria, and cause cell lysis in specific cases. This species-specific activity challenges the long-standing divide of antibiotics into bactericidal and bacteriostatic, and provides a possible explanation for the strong impact of macrolides on the gut microbiota composition in animals5-8 and humans9-11. To mitigate the collateral damage of macrolides and tetracyclines on gut commensals, we exploited the fact that drug combinations have species-specific outcomes in bacteria12 and sought marketed drugs, which could antagonize the activity of these antibiotics in abundant gut commensal species. By screening >1,000 drugs, we identified several such antidotes capable of protecting gut species from these antibiotics without compromising their activity against relevant pathogens. Altogether, this study broadens our understanding of antibiotic action on gut commensals, uncovers a previously unappreciated and broad bactericidal effect of prototypical bacteriostatic antibiotics on gut bacteria, and opens avenues for preventing the collateral damage caused by antibiotics on human gut commensals.
MAIN TEXT
Medication is emerging as major contributor for changes in the composition of the human gut microbiota4,13-15. Such severe and long-lasting changes are associated, and in some cases causatively linked, to dysbiosis and a wide range of diseases16. Although several non-antibiotic drugs may also have a previously unappreciated impact on the gut microbiome composition4,16,17, antibiotics, developed to have broad spectra and thereby target very diverse pathogens, are long known to take a heavy toll on our gut flora, causing a variety of gastrointestinal side-effects18, including Clostridioides (former Clostridium) difficile infections. Recently more attention has been given to this collateral damage of antibiotics on the gut microbiota and thereby on the host’s wellbeing. In vivo studies highlight links between the long-term microbiota compositional changes and host dysbiosis, including the development of allergic, metabolic, immunological and inflammatory diseases5-8,10,11,19-21. While uncovering the direct effects of different antibiotics on our gut flora is critical to improve general health, technical difficulties hamper routine testing of antibiotic susceptibility in anaerobes22,23. Currently available data on bacterial susceptibility to antibiotics is focused on invasive pathogens and offers little to no resolution in the diversity of the human gut microbiota24. Information is missing even for the most prevalent and abundant gut species, or ones recently associated with dysbiosis and disease25,26. In addition, existing animal or cohort studies have used a handful of antibiotics or merge data from different antibiotic classes, precluding systematic and general conclusions on the matter.
We recently assessed the direct effect of ∼1200 FDA-approved drugs on the growth of 38 prevalent and abundant or disease-associated human gut species under anaerobic conditions at a fixed concentration of 20 µM4. This initial screen (referred to hereafter as “screen”) included 144 antibiotics (Fig. 1a, Extended Data Fig. 1, Suppl. Table 1), with different classes having discernible effects on gut microbes (Fig. 1b). We validated these results by measuring 815 MICs (33 antibiotics and 2 antifungals for 17 species, 22 antibiotics for 10 additional species), using MIC gradient test strips (Fig. 1a, Extended Data Fig. 2, Suppl. Table 2 + 3). Despite differences in the experimental procedure, concordance between data from the initial screen and MICs is very high: Specificity and sensitivity of 0.97 (Extended Data Fig. 3a). The newly established MICs also correlate well with available data on antimicrobial susceptibility from databases such as EUCAST24 or ChEMBL27 (rs=0.69 and rs=0.64, respectively), despite differences in strains and media used (Extended Data Fig. 3b). Importantly, this new dataset considerably expands the available MICs, as much as by 80% for non-pathogenic bacteria (Fig. 1c, Extended Data Fig. 3c). Altogether, the initial screen and the new MIC dataset provide high-resolution information on the target spectrum of antibiotics on commensal gut microbes, which we explored further.
The antibiotics tested exhibited strong class-dependent effects (Fig. 1b, d). Consistent with literature, aminoglycosides hardly affected gut microbes under anaerobic conditions28 and sulfonamides were inactive in the medium used for the screen4. Quinolones acted in a generation-dependent manner. First-generation variants were effective only on a narrow spectrum of microbes that included both commensal E. coli tested. Second- and third-generation quinolones increased the spectrum. Fourth-generation variants (developed to increase activity against anaerobes) inhibited all tested species, except for Akkermansia muciniphila (Fig. 1e, Extended Data Fig. 1, red box), a species associated with protection against different diseases and dysbiotic states29, and even positive responses to immunotherapy30. For β-lactams, resistance was patchy but distinct for different members and subclasses (Extended Data Fig. 2, 4a). For Bacteroidetes, we tested additional species and strains (in total 12 and 19, respectively) (Extended Data Fig. 4b, c), confirming that β-lactam sensitivity and phylogenetic relatedness are uncoupled (Extended Data Fig. 4d). This argues for resistance mechanisms being strain-specific and horizontally transferred. Macrolides showed a strong impact on gut commensals and inhibited all tested microbes (Fig. 1d), except for the opportunistic pathogen C. difficile, which was resistant to all tested macrolides and clindamycin (Extended data Fig. 2, red box). This is in line with the associated risk of C. difficile infection after macrolide/clindamycin treatment31. Finally, 8 of the 9 tested tetracyclines inhibited nearly all tested microbes, which is surprising in the light of the gut microbiota being considered as reservoir for tetracycline resistance genes32.
Concentration-resolved MICs confirmed the same drug class-dependent trends observed in the screen (Fig. 1d, f). In addition, MICs allow for comparisons with clinical breakpoints, i.e. MICs at which a species should be considered resistant or susceptible (Fig. 1f). Overall, the gut microbes in our assays (anaerobic growth, gut mimetic growth medium33) were slightly more resistant to most antibiotic classes than previously reported for pathogens (aerobic growth, Mueller-Hinton agar). Tetracyclines were the exception, inhibiting commensals at significantly lower concentrations than what is reported for pathogens (Fig. 1f). Thus, commensals might be considerably less resistant to tetracyclines than previously anticipated and suggested by the detection of tetracycline resistance elements in fecal metagenomes.
Recent in-vivo studies have shown that β-lactams and macrolides have a strong and long-lasting collateral impact on the gut microbiota composition and thereby on host health5-8. As β-lactams exhibited strain-specific effects (Extended Data Fig. 1, 2, 4) and are known to kill bacteria (bactericidal), they could irrevocably deplete specific members of the gut microbiota, thereby explaining their differential and long-lasting effects on the community composition. On the other hand, macrolides uniformly targeted all tested gut commensals (Fig. 1d) and are textbook bacteriostatic antibiotics, i.e. inhibit bacterial growth, but do not kill (at least at high numbers). In this case, the long-term community composition change is more difficult to rationalize, as all community members are inhibited, but should be able to regrow once drug is removed. Similarly, tetracyclines, another class of bacteriostatic antibiotics that acted on nearly all gut microbes we tested, have known gastro-intestinal side-effects18, which are indicative of gut microbiome dysbiosis. We thus wondered at which level macrolides and tetracyclines exert a differential effect on gut microbes. Although traditionally both clinical use34-37 and basic research38,39 heavily rely on this distinction between bactericidal and bacteriostatic antibiotics, there are reports of drugs changing their killing capacity depending on the organism, drug concentration or medium tested40,41 (and increased evidence from meta-analyses that the distinction may have little relevance to clinical practice42,43). We therefore hypothesized that this bacteriostatic/bactericidal divide may be less rigid for gut commensals, which are more phylogenetically diverse than the few pathogens usually tested, and hence provide a level where the effect of these drug classes on gut microbes becomes differential.
The standard way to determine whether an antibiotic has bactericidal or bacteriostatic activity is to calculate time-kill curves, where the bacterial survivors are counted on agar at various time-points after drug treatment. If, over a significant period of antibiotic treatment (ranging from 5 to 24 hours), the number of colony forming units (CFU)/ml of culture decreases by more than 99.9%, the antibiotic is considered bactericidal40. We assessed the survival of 12 abundant gut microbes over a 5-hour treatment of either a macrolide (erythromycin or azithromycin) or a tetracycline (doxycycline) at 5 x MIC (Fig 2a + b, Extended Data Fig. 5). About half of the tested species decreased in survival by >99.9%, pointing to these drugs being bactericidal to several abundant gut microbes. To confirm this further, we tested the viability of B. vulgatus and E. coli ED1a upon erythromycin, azithromycin or doxycycline treatments using live/dead staining. Microscopy and flow cytometry assessment of live/dead bacteria corroborated the initial observations (Fig. 2c, Extended Data Fig. 6). As tetracyclines are considered bona-fide bacteriostatic drugs in E. coli, we were surprised to see that doxycycline effectively killed the commensal E. coli ED1a (Fig. 2a). We verified that these effects held also in the presence of oxygen (Extended Data Fig. 7a) and confirmed that doxycycline has a stronger bactericidal action on this natural isolate than on the domesticated E. coli K-12 lab strain, BW25113 (Extended Data Fig. 7b). In parallel, we excluded that the differences in killing capacity were confounded by growth rate, growth phase or MIC of the bacterial species tested (Extended Data Fig. 8). We also noticed that B. vulgatus and B. uniformis cultures decreased density in the presence of erythromycin (Fig. 2d). We confirmed by time-lapse microscopy that this was due to lysis. Erythromycin caused cell shape defects, including blebbing, cytoplasmic shrinkage, and ultimately cell lysis in both B. vulgatus and B. uniformis (Figure 2e, Movies 1-4). Altogether, this selective bactericidal activity of macrolides and tetracyclines on specific gut commensals could provide an explanation for the strong effects these drug classes have on the gut microbiota composition of human individuals. The gut microbes killed from the drug would be inadvertently removed from community, whereas the ones being only inhibited could recover when the therapy stops.
Knowing that drug combinations often have species-specific outcomes12, we reasoned that we could identify drugs that selectively antagonize the effect of antibiotics on gut microbes, while retaining activity against pathogens. Therefore, we screened the Prestwick library to identify antagonizing compounds to erythromycin or doxycycline on two abundant and prevalent gut microbes, B. vulgatus and B. uniformis (Fig. 3a, Extended Data Fig. 9). Of the 19 identified hits (Fig. 3b, Suppl. Table 4), we tested the 14 candidates with the strongest activity in a concentration-dependent manner (Extended Data Fig. 10a). Nine retained antagonistic activity over a broader concentration range, which we confirmed by checkerboard assays (Fig. 3c). The antidotes that showed the strongest antagonisms were the anticoagulant drug dicumarol, and two non-steroidal anti-inflammatory drugs, tolfenamic acid and diflunisal. While dicumarol rescued B. vulgatus from erythromycin and diflunisal from doxycycline, tolfenamic acid was able to protect B. vulgatus from both drugs. In addition, these interactions were able to at least partially rescue the killing of B. vulgatus by erythromycin and doxycycline (Extended Data Fig. 10b). We then probed two of these drugs for their ability to protect other abundant gut commensals and confirmed that both dicumarol and tolfenamic acid were able to counteract erythromycin on several species (Fig. 3d, Extended Data Fig. 11). In contrast, both drugs did not affect the potency of erythromycin on Staphylococcus aureus, Streptococcus pneumoniae and Enterococcus faecium, pathogens against which erythromycin is active/prescribed (Fig. 3e, Extended Data Fig. 12a). For example, tolfenamic acid and dicumarol at concentration ranges of 5-40 µM could rescue the growth of five out of seven tested abundant gut commensal species at clinically relevant erythromycin concentrations (Fig. 3f, Extended Data Fig. 12b). Altogether, our data provides a proof-of-principle for identifying antidotes that specifically mask the collateral damage of antibiotics on commensals. This concept would need to be further validated in the future in animal models. Antidotes may also need to be modified to late (colon)-release or non-absorbable formulations to ensure they reach the gut and to minimize adverse effects from their primary action.
In summary, our study provides a high-resolution map of the collateral damage of antibiotics on 50 different resident gut microbes down to the level of individual drugs, species and partially even strains. We challenge the universal divide of antibiotics into bacteriostatic and bactericidal across bacteria, as this breaks down when tested beyond model organisms. Antibiotics with preferential killing of some species may be the most detrimental to our gut flora, although the first studies in a few healthy individuals point to the gut microbiota having some resilience against specific antibiotic regimens44. Understanding the underlying mechanisms for this selective killing might open up ways for the development of new antimicrobials, but also strategies for controlled microbiome modulation15. Finally, we provide a proof-of-concept that species-specificity of drug combinations12 can be exploited to identify antidotes that selectively protect the gut microbiota from the adverse effects of systemic antibiotic therapy. This new approach adds to proposed and existing strategies of gut microbiota protection against antibiotics, such as co-administration of activated charcoal45, β-lactamases46, probiotics or (autologous) fecal transplants47. Overall, our results suggest that interactions of antibiotics and commensals merit deeper exploration, as our current knowledge of the mode(s) of action of antibiotics in model pathogens is not necessarily transferable to commensals.
METHODS
Growth conditions
All experiments from this study were performed in an anaerobic chamber (Coy Laboratory Products Inc) (2% H2, 12% CO2, 86% N2) and all materials and solutions used for these experiments were pre-reduced for at least 24 h before use unless specified otherwise. Bacteria used in this study were typically pre-cultured for two overnights: Cells were cultured in 5 ml modified Gifu Anaerobic Medium broth (MGAM) (HyServe GmbH & Co.KG, Germany, produced by Nissui Pharmaceuticals) and grown at 37°C overnight. The next day, cells were diluted 1/100 in 5 ml MGAM medium and grown at 37°C for a second overnight before starting the experiments.
Quantitative assay for minimum inhibitory concentration determination with MICs test strips
MICs test strips were purchased from Liofilchem or Oxoid (Suppl. Table 2). All MICs were measured under anaerobic growth conditions inside a Coy anaerobic chamber. Bacteria were precultured in MGAM for two overnights and cultures were diluted to OD578 = 0.5. 50 µl of the diluted culture were spread on a MGAM agar plate and allowed to dry for 15 min. The MIC test strip was placed on the agar with sterile tweezers, allowing the part with the lowest concentration touch the agar first. Plates were incubated at 37°C inside the anaerobic chamber, at least overnight and longer depending on the species-specific growth requirements. After formation of a symmetrical inhibition ellipse, plates were taken out of the chamber and imaged under controlled lighting conditions (spImager S&P Robotics Inc.) using an 18 megapixel Canon Rebel T3i (Canon Inc. USA). MICs were directly determined from the strip scale at the point where the edge of the inhibition ellipse intersects the MIC test strip. All MICs were determined in duplicates. In cases of an eight-fold difference between the two values, a third replicate was done. In all cases, this resulted in a clear outlier (> 8-fold different from other two MICs) that was removed from the dataset.
MIC comparison to ChEMBL and EUCAST databases
Previously known MICs were extracted from the ChEMBL database (version 24)27 and EUCAST (obtained on May 14, 2018)24. Antibiotics from these two datasets were mapped to our dataset by name. Species were mapped using NCBI Taxonomy Identifiers and species names. For MICs from ChEMBL, a keyword-based approach was used to exclude experiments on species with mutations, deletions, insertions, etc. The EUCAST database contains a large number of reported MICs for each compound–species pair. We collapsed these to a single value by calculating the median MIC.
Estimates on the abundance and prevalence of species in the healthy human gut microbiome were calculated using mOTUs v248 as follows: Relative species abundances were determined in 727 shotgun metagenomic samples from donors in the control groups of multiple studies from various countries and continents49-53. Prior to taxonomic profiling, metagenomes were quality controlled using the MOCAT2 -rtf procedure54, which removed reads with ≥95% sequence identity and an alignment length of ≥45bp to the human genome hg19. Taxonomic profiles were then created using mOTUs version 2.1.048 with parameters -l 75; -g 2; and -c. Afterwards relative abundances below 10−4 were set to zero and species with nonzero abundance in <5 samples discarded. For the retained 1,350 species, prevalence was defined as the percentage of samples with nonzero abundance; a prevalence cut-off of 1% was chosen to classify species into “rare” and “common” species. For all species in the MIC dataset, we manually assessed their status as pathogenic or non-pathogenic species using encyclopaedic and literature knowledge. Pathogenic species that occur in more than 1% of healthy people (i.e. are designated as “common”) were classified as “potentially pathogenic species” that can, for example, cause diseases in immunocompromised patients.
Killing curves and survival assay
Cells were precultured as described in the growth conditions section before being diluted to an OD578=0.01 and grown for 2 h at 37°C (unless specified otherwise). Next, cells were diluted 1/2 in MGAM containing a 10-fold MIC of erythromycin, azithromycin or doxycycline (final antibiotic concentration is 5-fold MIC) and incubated in the presence of the antibiotic for 5 h at 37°C. At several time-points (0, 1h, 2h, 3h, 4h, 5h), 100 μl of cells were serial-diluted in PBS (10−1 to 10−8 dilutions) and plated on MGAM-Agar plates for CFU counting. When no cells were detected using this method, a bigger volume of culture (up to 2 ml) was plated to be able to detect CFUs. Agar plates were incubated overnight at 37°C and colonies were counted the next day, either manually, for low CFU numbers, or using the Analyze Particles tool from ImageJ55.
Live/dead staining
Cells were precultured as described in the growth conditions section before being diluted to an OD578=0.01 and grown for 2 h at 37°C. Cells were next diluted 1/2 in MGAM containing 10-fold MIC of erythromycin, azithromycin or doxycycline (final concentration is 5-fold the MIC) and incubated in the presence of the antibiotic for 5 h at 37°C. Then, cells were live/dead stained using the LIVE/DEAD BacLight Bacterial viability and counting kit (#L34856 Molecular Probes, ThermoFisher) according to the manufacturer’s protocol before and after antibiotic treatment.
Flow cytometry
Stained cells were counted using a BD LSRFortessaTM flow cytometer. The forward and side scatter signals (488 nm) as well as the green and red fluorescent signals (488-530/30A filter and 561-610/20A filter, respectively) were acquired. The FSC/SSC detectors were set to logarithmic scale. The flow rate varied between 12 µl/min and 60 µl/min depending on the concentration of each sample, and the analysis was stopped when 10,000 target events were measured. Graphs were generated using the FlowJo V10.3 software (Treestar).
Microscopy
For live/dead imaging, stained cells were washed twice in 0.85% NaCl before being spotted on 0.85% NaCl +1% agarose pads between a glass slide and a coverslip. For time-lapse imaging, cells were precultured as described in the growth conditions section. Cells were then diluted to an OD578=0.01 and grown for 3 h at 37°C before being spotted on MGAM +1% agarose pads, supplemented or not with 15 µg/ml erythromycin (5-fold MIC) between a glass slide and a coverslip. Slides were sealed with valap (to avoid/delay oxygen permeation) and taken outside of the anaerobic chamber for imaging. In these conditions, untreated bacteria kept growing rapidly (Movie 1 + 3). The imaging was performed using a Nikon Eclipse Ti inverted microscope, equipped with a Nikon DS-Qi2 camera, a Nikon Plan Apo Lambda 60X oil Ph3 DM phase contrast objective and a Nikon HC mCherry filter set (Ex 562/40; DM 593; BA 641/75) to detect propidium iodide fluorescence. Images were acquired with the NIS-Elements AR4.50.00 software and processed with Fiji v.2.0.0-rc-68/1.52h56.
Growth curves
Cells were precultured as described in the growth conditions section. Then, cells were diluted to an OD578=0.01 in a 96-well plate sealed with a breathable membrane (Breathe-Easy®) and grown for 2 h. Next, erythromycin was added to the culture to a final concentration of 15 µg/ml (5-fold MIC) and growth curves were acquired for 20 h using a microplate spectrophotometer (EON, Biotek) by measuring the OD578 every hour after 30 sec of linear shaking.
Screen for microbiome-protective antibiotic antagonism
Preparation of screening plates
The Prestwick Chemical Library was purchased from Prestwick Chemical Inc. and drugs were re-arrayed, diluted and stored in 96 well format as described before4. We prepared drug plates (2 x drug concentration) in MGAM medium and stored them at -30°C. For each experiment, drug plates were thawed, supplemented with the respective antibiotic solution (freshly prepared in MGAM) and pre-reduced in the anaerobic chamber overnight. All rearranging and aliquoting steps were done using the Biomek FXP (Beckman Coulter) system.
Inoculation and screening conditions
Strains were grown twice overnight, the second overnight culture was diluted in MGAM to reach OD578 nm 0.04 (4 x the desired starting OD). 25 µl of the diluted cultures were used to inoculate wells containing 50 µl of 2x concentrated Prestwick drug and 25 µl of the 4x concentrated antibiotic using the semi-automated, 96-well multi-channel pipette epMotion96 (Eppendorf). Each well contained 1% DMSO, 20 µM of the Prestwick drug and a species-specific antibiotic concentration that was just inhibitory for the respective strain (0.625 µM for erythromycin, 0.04 µM doxycycline for B. uniformis and 0.08 µM doxycycline for B. vulgatus). Plates were sealed with breathable membranes (Breathe-Easy®) and OD578 was measured hourly after 30 sec of linear shaking with a microplate spectrophotometer (EON, Biotek) and an automated microplate stacker (Biostack 4, Biotek) fitted inside a custom-made incubator (EMBL Mechanical Workshop). Growth curves were collected up to 24 h. For each antibiotic, each species was screen in biological duplicates. All experiments included control wells of unperturbed growth (32 wells per run) and control wells for growth in the presence of the antibiotic only (8 wells per plate).
Analysis pipeline and hit calling
All growth curves within a plate were truncated at the transition time from exponential to stationary phase and converted to normalized AUCs using in-run control wells (no drug) as described before4. We then calculated z-scores based on these normalized AUCs, removed replicates with 8-fold differences in z-scores to eliminate noise effects, computed mean z-scores across the two replicates and selected combinations with mean z-scores > 3. This selection included 19 potential antibiotic antagonists and we followed up on 14 of them (7 potential erythromycin and 7 potential doxycycline antagonists in either B. vulgatus or B. uniformis – see Extended Data Fig. 9) in independent experiments. Validation of microbiome-protective antagonists. First, we kept the erythromycin/doxycycline concentration constant (0.625 µM for erythromycin, 0.078 µM (B. vulgatus)/ 0.039 µM (B. uniformis) for doxycycline) and tested concentration gradients of the potential antagonists with ranges depending on the antagonist’s solubility. Compounds were purchased from independent vendors (Suppl. Table 5) and dissolved at 100x starting concentration in DMSO. Eight 2-fold serial dilutions were prepared in 96-well plates with each row containing a different antagonist, sufficient control DMSO wells and wells with just the respective antibiotic (‘antibiotic-only’ control). These master plates were diluted in MGAM medium (50 µl) to 2 x assay concentration and 25 µl freshly prepared antibiotic solution (4x test concentration) was added. Plates were pre-reduced overnight in an anaerobic chamber and inoculated with 25 µl of overnight cultures (prepared as described under Growth conditions) to reach a starting OD578 of 0.01 and 1% DMSO concentration. Growth was monitored hourly for 24 h after 30 sec of linear shaking (as described for the screen4). Experiments were performed in biological triplicates. For analysis, growth curves were converted into normalized AUCs (see above). We accounted for residual growth in the presence of the antibiotic by subtracting the median normalized AUCs of the ‘antibiotic-only’ control per plate. We computed medians across triplicates and considered a normalized AUC > 0.25 as concentration-dependent growth rescue by the antagonist.
Checkerboard assays for anaerobic commensals
Validated antagonists were further investigated in 8×8 checkerboard assays, where both antibiotics and antagonists were titrated against each other. Such assays were first performed for the commensals that were originally screened (i. e. B. vulgatus and B. uniformis – 4 replicates) and later expanded towards six further gut microbes (B. caccae, B. fragilis NT, B. ovatus, B. thetaiotaomicron, P. copri, P. distasonis – 2 replicates). For vertical gradients, 2-fold serial dilutions of the antagonists were prepared first in 100x in DMSO and diluted in MGAM as described above (section ‘Validation of microbiome-protective antagonists’). Horizontal antibiotic dilution series were freshly prepared in MGAM at 4x final concentration in equidistant concentration steps. Both, vertical and horizontal dilution series were combined (50 µl of the antagonist gradients (2x) and 25 µl of the antibiotic gradients (4x)) in 96 well plates. Plates were pre-reduced under anaerobic conditions overnight, inoculated with 25 µl of diluted overnight culture (at 4x starting OD) and sealed with breathable membrane (Breathe-Easy®). Bacterial growth was monitored once per hour for 24 h after 30 sec linear shaking (Eon + Biostack 4, Biotek) under anaerobic conditions. Growth curves were converted into normalized AUCs as described using in-plate controls to define unperturbed growth.
Checkerboard assays for pathogens under aerobic conditions
For three pathogens (S. aureus DSM20231 ATCC 12600 and E. faecium ATCC19434) 8×8 checkerboard assays were performed in transparent 384 well plates (Greiner BioOne GmbH), with each well containing a total volume of 30 µl in total for S. aureus and 55 µl for E. faecium. S. aureus strains were grown in tryptic soy broth (TSB, Sigma Aldrich), E. faecium in BHI medium (Sigma Aldrich). Drugs were arrayed in 2-fold serial dilutions for the checkerboards. Cell were inoculated at initial OD595nm ∼0.01 from an overnight culture. Plates were sealed with breathable membranes (Breathe-Easy), incubated at 37°C (Cytomat 2, Thermo Scientific) with continuous shaking and OD595nm was measured every 30 min for 16 h in a Filtermax F5 multimode plate reader (Molecular Devices). For S. pneumoniae D39, we only tested concentration gradients of the potential antagonists in a constant antibiotic concentration (0.2 µM erythromycin) in BHI medium. All experiments were done at least in 2 biological replicates and 2 technical replicates. Wells in which there was significant condensation were removed and background due to medium was subtracted. Growth curves were trimmed at the transition to stationary phase (9 h for S. aureus, 12 h for E. faecium). AUCs were calculated and normalised by the median of the internal no-drug control wells (n = 6). Interactions were quantified according to the Bliss interaction model57. Interactions were called antagonistic if the median of all the interaction scores for a given checkerboard was above 0.05, synergistic if the value was below -0.05 and neutral if lying between these two cut-offs.
Phylogenetic analysis/phylogenetic tree construction
In order to generate a phylogenetic tree for the different isolates, the nucleotide sequences for a set of universally occurring, protein coding, single copy phylogenetic marker genes48,58 were extracted from reference genomes or genome assemblies using fetchMG58 (https://motu-tool.org/fetchMG.html). Within the framework of the ete3 toolkit59, ClustalOmega60 was used to create sequence alignments for each marker gene independently and all columns with more than 10% gaps were removed. The individual alignments were concatenated and finally, a phylogenetic tree was inferred from the combined alignment using IQTree61.
Data availability
Data is available upon request.
Code availability
Code is available upon request.
SUPPLEMENTARY INFORMATION
Supplementary tables
Supplementary Table 1
Annotation and classification of antibiotics
Supplementary Table 2
MICs strips used in this study
Supplementary Table 3
MICs as determined by MIC strips for selected antibiotics
Supplementary Table 4
Hitlist: Screen for microbiome-protective antagonists to erythromycin/tetracycline
Supplementary Table 5
Compounds used in this study
Supplementary movies
Movie 1
Time-lapse of B. vulgatus growing on MGAM-Agarose 1% pad.
Movie 2
Time-lapse of B. vulgatus growing on MGAM-Agarose 1% pad containing 5-fold MIC of erythromycin.
Movie 3
Time-lapse of B. uniformis growing on MGAM-Agarose 1% pad.
Movie 4
Time-lapse of B. uniformis growing on MGAM-Agarose 1% pad containing 5-fold MIC of erythromycin.
Author contributions
This study was conceived by KRP, PB and AT; designed by LM, CVG, MP and AT; and supervised by LM and AT. Experiments were conducted by LM, CVG and MP. TB and EEA contributed to MIC measurements and EC to checkerboard analyses. Data preprocessing, curation and comparisons to existing databases were performed by JW, MK, AM, UL, SKF and GZ. Data interpretation was performed by LM, CVG, JW, GZ and AT. LM, CVG and AT wrote the manuscript with feedback from all authors; LM, CVG, JW and MK designed figures with inputs from GZ and AT. All authors approved the final version for publication.
Author information
EMBL has filed a patent application on using the antidotes identified in this study for prevention and/or treatment of dysbiosis and for microbiome protection (European patent application number EP19216548.8). LM, CVG, EC and AT are listed as inventors. Correspondence and requests for materials should be addressed to typas{at}embl.de or l.maier{at}uni-tuebingen.de.
Acknowledgements
We thank Stephan Göttig and the Typas lab for feedback on the manuscript. We thank Ana Rita Brochado, Sarela Santamarina and the EMBL Flow Cytometry Core Facility for help with experimental design. We acknowledge EMBL and the JPIAMR grant Combinatorials for funding. LM and MP were supported by the EMBL Interdisciplinary Postdoc (EIPOD) program under Marie Sklodowska Curie Actions COFUND (grant 291772) and LM is now supported by the CMFI Cluster of Excellence (EXC 2124). CG is recipient of an EMBO long-term postdoctoral fellowship and an add-on fellowship from the Christiane Nüsslein-Volhard-Stiftung. AT is supported by ERC consolidator grant uCARE.