A Pathogen-Selective Antibiotic Minimizes Disturbance to the Microbiome

Abstract

Broad-spectrum antibiotic therapy decimates the gut microbiome, resulting in a variety of negative health consequences. Debio 1452 is a staphylococcus-selective enoyl-acyl carrier protein reductase (FabI) inhibitor under clinical development and was used to determine whether treatment with pathogen-selective antibiotics would minimize disturbance to the microbiome. The effect of oral Debio 1452 on the microbiota of mice was compared to the effects of four commonly used broad-spectrum oral antibiotics. During the 10 days of oral Debio 1452 treatment, there was minimal disturbance to the gut bacterial abundance and composition, with only the unclassified S24-7 taxon reduced at days 6 and 10. In comparison, broad-spectrum oral antibiotics caused ∼100- to 4,000-fold decreases in gut bacterial abundance and severely altered the microbial composition. The gut bacterial abundance and composition of Debio 1452-treated mice were indistinguishable from those of untreated mice 2 days after the antibiotic treatment was stopped. In contrast, the bacterial abundance in broad-spectrum-antibiotic-treated mice took up to 7 days to recover, and the gut composition of the broad-spectrum-antibiotic-treated mice remained different from that of the control group 20 days after the cessation of antibiotic treatment. These results illustrate that a pathogen-selective approach to antibiotic development will minimize disturbance to the gut microbiome.

INTRODUCTION

The discovery and commercialization of broad-spectrum antibiotics is one of the most significant medical achievements in the first half of the 20th century. However, the use of broad-spectrum antibiotics is linked to a variety of medical problems, such as antibiotic-treatment-associated infections first recognized in the 1950s (1, 2). Broad-spectrum-antibiotic therapy is known to devastate the gut microbiome, and the repeated use of these antibiotics during early childhood is linked to metabolic and autoimmune diseases later in life (3–10). Therefore, one challenge for future antibacterial therapeutic development is to design drugs that minimize disturbances to the microbiome.

One approach to minimize disturbance to the microbiome is to use narrow-spectrum, pathogen-selective antibiotics. In principle, antibiotics optimized to target a single pathogen would not impact the beneficial inhabitants of the gut. There are two ways to achieve pathogen selectivity. First, design an inhibitor against an enzyme or process that is only found in the targeted pathogen. This approach has the challenge of identifying a novel, essential target in each pathogen. A second approach circumvents this issue and utilizes structure-based design to build high-affinity inhibitors that are optimized against the specific version of a drug target expressed in the pathogen. One antimicrobial target that has been exploited for the development of a pathogen-selective inhibitor is enoyl-acyl carrier protein (ACP) reductase (FabI) (11–16). Debio 1452, previously known as AFN-1252 (Fig. 1), is an example of a FabI inhibitor with the properties of a pathogen-selective antibiotic based on both target distribution and isozyme selectivity (17). Debio 1452 is specifically designed to target staphylococcal FabI, a key component of bacterial fatty acid synthesis (18–20). FabI is essential in Staphylococcus species (11) but is not found in many other groups of bacteria. Specifically, members of the order Clostridiales are abundant residents of the gut, and free-living members of this group do not encode a FabI but, rather, utilize an unrelated enoyl-ACP reductase called FabK (21, 22). Debio 1452 is also highly selective for staphylococcal FabI compared to the essential FabIs expressed in other bacteria, due to a specific drug interaction with an active-site methionine that is unique to staphylococcal FabI (12, 23). Consequently, Debio 1452 has potent activity against Staphylococcus aureus and other staphylococcal spp., including multidrug-resistant isolates, but does not inhibit the growth of bacteria of many other Gram-positive or Gram-negative genera (24). These considerations suggest that Debio 1452 therapy may have minimal impact on the gut microbiome. However, experimental validation of this idea is essential because most of the gut inhabitants are known only from their 16S rRNA gene sequences, and whether or not they express an essential FabI with an S. aureus-like active site is unknown. In this study, we examined the effect of therapeutic doses of Debio 1452 on the gut microbiome of mice in comparison to the effects of a panel of broad-spectrum antibiotics. As anticipated (2, 25, 26), the broad-spectrum agents devastated the gut microbiome. However, Debio 1452 treatment caused minimal effects on both the bacterial abundance and the composition of the gut microbiome, illustrating that pathogen-selective antibiotics can be developed to minimize disturbances to the microbiome.

FIG 1.

Structure of Debio 1452. AFN-1252 was recently acquired by Debiopharm International SA and is referred to clinically as Debio 1452.

MATERIALS AND METHODS

Mouse husbandry.

Animal experiments were approved by the St. Jude Children's Research Hospital Institutional Animal Care and Use Committee (protocol 538). Mice were housed in the St. Jude Children's Research Hospital Animal Research Center at 22 ± 2°C, 50% ± 10% humidity, and a 14-h light/10-h dark cycle. The animals were fed Purina Diet 5 chow. Six-week-old female C57-BL6 mice were ordered from Jackson Laboratory. The mice were handled, received abdominal massage, and were mock gavaged twice weekly during the acclimation period. Stool samples were collected from each mouse via abdominal massage before antibiotic treatment started, daily during a 10-day antibiotic treatment, and on days 12, 17, 23, 30, and 37 during the postantibiotic recovery. Stool samples were stored at −80°C until extraction. Antibiotics were purchased from Santa Cruz Biotechnology, except for Debio 1452, which was provided by Debiopharm International SA. The antibiotic stock solutions were formulated in 1% Pluronic F-127 (Sigma-Aldrich) at 29.4 mg/ml for Debio 1452 tosylate monohydrate salt, 40 mg/ml for linezolid, 40 mg/ml for clindamycin, 13 mg/ml for moxifloxacin, and 48 mg/ml for amoxicillin. Antibiotic treatments were given once daily via oral gavage with the stock antibiotic during the antibiotic treatment phase. The dose levels of broad-spectrum antibiotics were based on the conversion of the human doses to body surface area (27) and were as follows: linezolid, 200 mg/kg of body weight; clindamycin, 200 mg/kg; moxifloxacin, 65 mg/kg; and amoxicillin, 240 mg/kg. The Debio 1452 dosage of 100 mg/kg was based on previous mouse studies (28). The drug carrier only (1% Pluronic F-127) was given to the control group.

DNA extraction.

Total DNA was extracted from 50 to 200 mg of stool sample using the QIAamp fast DNA stool minikit. InhibitEX buffer (1 ml) was added to the stool sample in an MP FastPrep tube. The tube was heated at 70°C for 10 min, and then the mixture was homogenized via shaking in the MP FastPrep-24 machine (4.0 m/s for 20 s 3 times). The tube was heated at 70°C for 10 more minutes and centrifuged to pellet stool particles. The rest of the extraction process followed the manufacturer's protocols. Stool samples from days −1, 2, 6, 10, 12, 17, 23, 30, and 37 were extracted for DNA and analyzed.

16S rRNA gene abundance determination.

Real-time PCR using Applied Biosystems SYBR green PCR master mix in the Applied Biosystems 7500 real-time PCR system was used to determine the relative abundances of 16S rRNA genes in extracted DNA samples. A 20-μl reaction mixture composed of 10 μl 2× SYBR green PCR master mix, 150 nM each forward and reverse DNA primers, and 5 μl of different concentrations of extracted DNA was amplified for 40 cycles. The 16S PCR primers were 5′-TCCTACGGGAGGCAGCAGT and 5′-GGACTACCAGGGTATCTAATCCTGTT (29). A serial dilution of DNA extracted from the stool of mice before antibiotic treatment was tested to determine the DNA dilution that gave the best dynamic range (cycle threshold values between 15 and 20). A 1-to-2,000 dilution of the extracted DNA was determined to give the best dynamic range for 16S rRNA gene detection and used to determine the cycle threshold for each of the samples. The resulting data were plotted in log₂ scale as the relative abundance for each mouse in each treatment group.

The abundance of the mouse tumor necrosis factor alpha (TNF-α) gene (primers 5′-GGCTTTCCGAATTCACTGGAG and 5′-CCCCGGCCTTCCAAATAAA) was also measured in the samples by real-time PCR as a control to monitor the efficiency of the extraction (30). A 1-to-10 dilution of the extracted DNA was used to determine the cycle threshold for each sample. The cycle threshold values for the TNF-α had a bell-shaped distribution with a mean of 30.14, median of 29.96, and range of 24 to 36. This normal distribution was reflected by a similar efficiency of DNA extraction across the samples.

16S rRNA gene sequencing.

The V1-to-V3 (V1-V3) region 16S rRNA gene amplicon library was generated via PCR using the NEXTflex 16S V1-V3 amplicon-seq kit (Bioo Scientific), following the manufacturer's instructions. The DNA from the PCR was cleaned using the Ampure XP PCR purification kit (Beckman Coulter). The PCR products were quantified using the Quant-iT PicoGreen assay (Illumina) and normalized by DNA concentration for sequencing. The samples were analyzed via paired-end sequencing using the Illumina MiSeq platform, following the manufacturer's protocols. The broad-spectrum-antibiotic treatments caused severe reductions in the numbers of read counts obtained from next-generation sequencing, corresponding to the reductions in 16S rRNA gene abundance. These samples were analyzed using the same methodology, because the reductions of counts represented real changes in the gut microbiome composition. The sequencing read counts were 100-fold lower than the mean for the respective cohort for Debio 1452 treatment mouse 5 on day 6, clindamycin treatment mouse 3 on day 2, and moxifloxacin treatment mouse 5 on day 37. These samples were considered sequencing failures and excluded from the analysis.

Taxon assignments.

The 16S primers targeting the V1-V3 regions were aligned to the full set of 16S rRNA gene sequences from the Greengenes database version 13.5 (31) using exonerate (32). Each 16S rRNA gene database sequence was truncated to include only the V1-V3 region, the primer-matching regions, and an additional 40 bases on either side. Duplicate V1-V3 regions were removed from the data set to create a unique V1-V3 representative reference set. All taxon labels from the duplicates that were removed were associated with the matching remaining representative sequences. Reads from each sample were aligned exhaustively to the nonredundant V1-V3 reference set using USEARCH (33), allowing a minimum sequence identity of 97%. All taxon labels associated with all top-scoring sequences were used to determine the taxon assignment of each read. The highest-resolution nonconflicting taxon of all taxa associated with the top-scoring V1-V3 region(s) was assigned as the taxon for a read.

Composition analysis.

The phylogenetic composition was analyzed in R version 3.1.1 using the package phyloseq version 1.8.2 (34). The taxa assigned for each sample were tabulated and converted to the percentage of the composition for each taxon. The tabulated count number data for each sample are organized by experiment day and treatment group and are included in Dataset S1 in the supplemental material. The resulting tables for all samples were converted into a phyloseq class. Distance plots for the samples were generated using the ordinate function with multidimensional scaling (“MDS”) as the method choice and betadiversity measure 1 (“w”) as the distance choice. The resulting principal components 1 and 2 were plotted as 2-dimensional distance plots grouped by treatment day. Bar charts were generated showing the family-level taxon composition for each treatment group for each treatment day by averaging the percent composition of the different samples in each treatment group.

RESULTS AND DISCUSSION

Gut bacterial abundance and composition.

Groups of 5 mice were treated orally with drug carrier only (control), Debio 1452, linezolid, clindamycin, moxifloxacin, or amoxicillin for 10 days and then allowed to recover for 27 additional days. Treatment with Debio 1452, linezolid, moxifloxacin, and amoxicillin were well tolerated with no notable distress or weight loss. Four mice in the clindamycin treatment group were distressed during the first 4 days; two recovered, but two mice died on days 4 and 5. Feces from the mice were collected 1 day before treatment and over the course of the experiment, and the total DNA was extracted from the stool samples (Fig. 2). The relative bacterial abundances were determined by performing real-time PCR amplifying the 16S rRNA gene from the total DNA (Fig. 3 and 4). The gut microbiome composition was determined through next-generation sequencing of the V1-V3 region of the 16S rRNA gene from the total DNA. The gut microbiome compositions of all the samples were compared using beta-diversity and multidimensional scaling analysis and plotted as 2-dimensional distance plots of the first two principal components (Fig. 3 and 4). In these distance plots, samples that are clustered together are more similar, while samples that are more spatially distant are more different. The gut compositions were also plotted as bar charts at the family taxonomic level (Fig. 5, 6, and 7). Distance plots are used to provide a high level of visualization of the similarities and differences in complex samples, whereas the bar charts identify the distribution of taxa.

FIG 2.

Timeline of stool samples collected for analysis.

FIG 3.

Relative 16S rRNA gene abundances (top plot in each panel) and distance analysis of beta-diversity measures between the different antibiotic treatment groups (bottom plot in each panel) on day −1 prior to therapy and during 10 days of therapy. Bars are averages and whiskers are standard errors of the mean.

FIG 4.

Relative 16S rRNA gene abundances (top plot in each panel) and distance plot in each panel of beta-diversity measures between the different antibiotic treatment groups (bottom plot in each panel) during the recovery from therapy. Bars are averages and whiskers are standard errors of the mean.

FIG 5.

Family-level distribution of bacterial taxa on day −1, prior to therapy, and during the 10 days of therapy. The top plot in each panel contains the high-abundance taxa, and the bottom plot in each panel contains the low-abundance taxa. Some sequences could not be assigned at the family level and were assigned at the order level. These taxa are denoted with square brackets around the assignment (e.g., [Clostridiales]).

FIG 6.

Family-level distribution of bacterial taxa during recovery from therapy. The top plot in each panel contains the high-abundance taxa, and the bottom plot in each panel contains the low-abundance taxa. Some sequences could not be assigned at the family level and were assigned at the order level. These taxa are denoted with square brackets around the assignment (e.g., [Clostridiales]).

FIG 7.

Family-level distribution of bacterial taxa 37 days after cessation of therapy. The top plot illustrates the composition of the high-abundance taxa, and the bottom plot shows the distribution of low-abundance taxa at day 37 posttherapy. Some sequences could not be assigned at the family level and were assigned at the order level. These taxa are denoted with square brackets around the assignment (e.g., [Clostridiales]).

The mice from different treatment groups had similar bacterial abundances before drug treatment. The samples from the different treatment groups obtained prior to treatment clustered closely together in the distance analysis, showing that the bacterial compositions in all the mice were similar before treatment. In terms of composition, the Rikenellaceae and S24-7 families (both of the Bacteroidia class) made up approximately 80% of the microbiome composition before treatment in all groups and in the untreated mice over the course of the experiment. The order Clostridiales, including Ruminococcaceae, Lachnospiraceae, Dehalobacteriaceae, Clostridiaceae, Mogibacteriaceae, and unclassified members, made up 5 to 15% of the microbiome composition. Akkermansia muciniphila of the Verrucomicrobia phylum averaged 1% of the microbiome composition. The RF39, Anaeroplasmataceae, Lactobacillaceae, Turicibacteraceae, Coriobacteriaceae, and Erysipelotrichaceae families were minor taxa making up 0.01 to 1% of the composition. The percent composition of individual taxa in the control mice were observed to vary over the course of the experiment and between the individual mice, but the overall patterns of the taxa in the gut of untreated mice were similar over the entire course of the experiment.

Debio 1452 treatment caused minimal disturbance to the gut microbiome.

Debio 1452 treatment did not cause a significant change in the bacterial abundance over the course of the 10-day treatment (Fig. 3) or the subsequent 27-day recovery (Fig. 4). Distance analysis showed that Debio 1452-treated mice clustered with the untreated mice over the course of the experiment, consistent with minimal changes in the composition of the gut microbiome (Fig. 3 and 4). Only one taxon, the S24-7 family, decreased following Debio 1452 treatment. The S24-7 family, which comprised 30 to 50% of the bacteria before treatment, was reduced to an average of 4% by day 6 and constituted only 0.2% of the bacteria by day 10 (Fig. 5). No other taxa had compositional decreases, and several taxa, including the Rikenellaceae family, the Clostridiales order, and A. muciniphila (Verrucomicrobiales order), exhibited compensating compositional increases to maintain the bacterial abundance during the decline in S24-7 (Fig. 5). The drug treatments were stopped after day 10, and the microbiome composition in the Debio 1452-treated mice was evaluated on day 12. By day 12, the S24-7 family had recovered to pretreatment levels in the Debio 1452-treated mice, and the gut composition of Debio 1452-treated mice was indistinguishable from that of untreated mice during the entire recovery phase (days 12 to 37), showing that this FabI inhibitor did not have a lasting effect on the microbiome (Fig. 4 and 6).

Linezolid, clindamycin, and amoxicillin severely perturbed the gut microbiome.

Linezolid, clindamycin, and amoxicillin treatment all caused an ∼4,000-fold decrease in the gut bacterial abundance at the second day of treatment that persisted until the end of the treatment (Fig. 3). Distance plot analyses showed that the bacterial compositions of linezolid-, clindamycin-, and amoxicillin-treated mice became significantly distant from that of the control group during therapy, consistent with the microbiome compositions becoming severely altered (Fig. 3). In each case, the remaining bacterial taxa were unique to the antibiotic used and represented those bacteria that survived the antibiotic treatment (Fig. 5). However, the persistent low bacterial abundance over the course of treatment means that the remaining bacterial taxa did not repopulate the gut microbiome (Fig. 3). The major taxon in amoxicillin-treated mice at days 6 and 10 was the unclassified order Clostridiales, suggesting that the bacteria from this order are relatively more resistant to amoxicillin treatment than the other bacteria in the gut microbiome (Fig. 5). The major taxon in clindamycin-treated mice at days 6 and 10 was the Pseudomonadaceae family, suggesting that bacteria of this family are better equipped than others to withstand clindamycin treatment (Fig. 5). The major taxa in linezolid-treated mice had a distribution similar to that in the untreated mice, suggesting that linezolid was equally effective at eliminating all of the significant taxa of the gut microbiome (Fig. 5).

Although the drug treatments were stopped at day 10, the bacterial abundance did not recover until day 17 for the linezolid and amoxicillin treatment groups and day 23 for the clindamycin treatment group (Fig. 4). The distance plots show that the bacterial compositions became distally closer to that of the control group over the recovery phase, but those of the clindamycin and amoxicillin treatment groups still did not cluster completely with the control group by day 37 (Fig. 4). For all three treatment groups, the Enterococcaceae family became a major taxon at day 12 (Fig. 6). On the first day when the bacterial abundance reached normal levels in each of these cases (Fig. 4), the S24-7 family became the major compositional component of the gut microbiome, making up 80 to 90% of the composition (Fig. 6). These data illustrate that the S24-7 family was the first normally observed taxon to recover in the gut microbiome following broad-spectrum antibiotic therapy in mice. By day 37, the major taxa had recovered to normal levels in all three cases. However, the minor taxa were still significantly different from those of the untreated mice in the clindamycin and amoxicillin treatment groups (Fig. 4, 6, and 7).

Moxifloxacin-specific microbiome alterations.

Of the broad-spectrum antibiotics tested, moxifloxacin treatment caused the least reduction in the abundance of bacteria in the gut microbiome (Fig. 3). Gut bacterial abundance was still significantly reduced, but only 16- to 100-fold over the course of the moxifloxacin treatment, compared to ∼4,000-fold by the other broad-spectrum antibiotics. Distance plot analysis showed that the microbiome composition in the moxifloxacin-treated mice became significantly distinct from that of the control group (Fig. 5). The percent composition of A. muciniphila increased over the course of the moxifloxacin treatment from 0.5% to 2% before treatment to 30% during day 2, 40% at day 6, and 50% at day 10 of treatment. The 100-fold increase in A. muciniphila at day 10 of moxifloxacin therapy was not due to an increase in this species but was rather due to the 100-fold decrease in the other species. The normal taxa constituted the other 50% of the microbiome composition at day 10, showing that moxifloxacin did not reduce the normal gut inhabitants as much as the other broad-spectrum antibiotics tested. The gut bacterial abundance of moxifloxacin-treated mice recovered 2 days after the treatment stopped, faster than for the other broad-spectrum antibiotics tested (Fig. 4). The pattern of compositional recovery was similar to the patterns in mice treated with other broad-spectrum antibiotics, with S24-7 recovering first, the other major taxa returning to normal by day 30, and the minor taxa remaining unstable and continuing to fluctuate at day 37.

Pathogen-selective antibiotic strategy minimizes disturbances to the microbiome.

One of the challenges of microbiome studies is applying mouse results to humans. The drug dosages used in this study were based on the equivalent surface area dosage conversion method to achieve similar plasma levels of drugs (27). Mice received 12 times the drug per mass compared to humans under this established conversion. Therefore, the concentration of the drug in the mouse gut was higher than would be found in the human gut, facilitating a rigorous test of Debio 1452 and the pathogen-selective approach for antibiotic discovery. The minimal perturbation of the gut microbiome by Debio 1452 therapy illustrates that a pathogen-selective antibiotic strategy is an effective approach to minimize disturbance to the gut microbiome. Debio 1452 derives its pathogen selectivity by targeting an enzyme that is either absent or nonessential in many important residents of the gut and by targeting the staphylococcal FabI selectively over other FabI homologues. Debio 1452 did not change the gut bacterial abundance over the course of the treatment. The only taxon reduced by Debio 1452 was the S24-7 family, which was replaced by an increase in the remaining taxa. As reported previously (2, 25, 26), the broad-spectrum antibiotics in our study caused a 100- to 4,000-fold reduction in gut bacterial abundance and severely altered the microbiome taxon composition. The gut composition of Debio 1452-treated mice was indistinguishable from that of untreated mice 2 days after treatment had ceased, while the gut compositions of broad-spectrum-antibiotic-treated mice took over 27 days to recover. Developing therapeutic strategies that minimize disturbances to the beneficial microbiome bacteria will become a more important criterion in antibiotic drug design in light of the emerging understanding of how the gut microbiome supports health by facilitating digestion, shielding us from invading microorganisms, and providing vitamins (3).

The principle inhabitants of the gut microbiome can bypass FabI inhibition.

The results of this study provide insight into the bacterial microbiome taxa that encode an essential FabI. The essentiality of FabI and de novo fatty acid synthesis is well characterized in a variety of human pathogens (11, 35, 36) but is largely unknown for the bacterial constituents of the gut microbiome. Known members of the Clostridiales order encode a FabK rather than a FabI (21, 22), and our results support the prediction that the Clostridiales in the gut are unaffected by FabI inhibition. It is more difficult to predict the behavior of the Bacteroidia class of organisms because FabI and FabK are both found in organisms belonging to this class. Based on sequenced genomes of family members (NCBI Microbial Genomes Resources), members of the Prevotellaceae family encode only a FabI, members of the Porphyromonadaceae family encode only a FabK (37), and the Bacteroidaceae and Rikenellaceae families encode both FabI and FabK. The observation that Debio 1452 does not reduce the abundance of the Rikenellaceae family, whose members are predicted to encode both FabI and FabK, indicates that the FabK enoyl-ACP reductase compensates for a Debio 1452-inactivated FabI. Alternately, Debio 1452 may be a poor inhibitor of Rikenellaceae FabI or this family may have a pathway to bypass fatty acid synthesis by obtaining fatty acids from the gut. The major impact of Debio 1452 therapy was the slow disappearance of the unclassified S24-7 family of the class Bacteroidia. This family was replaced by the expansion of the Clostridiales, but S24-7 returned to normal levels on day 2 following the cessation of treatment. These data suggest that the S24-7 family has an essential FabI that is not potently inhibited by Debio 1452. Genomic data are unavailable for the S24-7 family, but the weak sensitivity to Debio 1452 may indicate that some Bacteroidia families may be affected by FabI inhibitors. S24-7 is not a component of the human gut. Instead, the major Bacteroidia species of the human microbiome belong to the Prevotellaceae and Bacteroidaceae families. Whether the Prevotellaceae have a FabI that is sensitive to Debio 1452 remains to be established. A. muciniphila of the Verrucomicrobia phylum is a minor but key component of the human gut microbiome (38). A. muciniphila is predicted to possess a FabI with high homology to the S. aureus FabI (WP_012420677.1, E value of 2e−71) and does not harbor any known alternative enoyl-ACP reductases. However, the abundance of A. muciniphila was not reduced by Debio 1452 therapy, suggesting that this organism has mechanisms to prevent the intracellular accumulation of Debio 1452, expresses an unknown, alternate enoyl-ACP reductase, or bypasses fatty acid synthesis inhibition by incorporating gut fatty acids derived from the diet. The rare taxa of the gut microbiome, such as the Fusobacteria, Actinobacteria, and Mollicutes, did not occur with sufficient frequency to draw conclusions regarding their FabI status. Understanding fatty acid metabolism in the gut bacteria would advance our knowledge of how FabI therapeutics, and fatty acid synthesis inhibitors in general, may impact the microbiome.

Funding Statement

This work was also funded in part by a sponsored research agreement with Debiopharm International SA.