Infection and antibiotic-associated changes in the fecal microbiota of C. rodentium ϕstx2_dact-infected C57BL/6 mice

ABSTRACT

Enterohemorrhagic Escherichia coli causes watery to bloody diarrhea, which may progress to hemorrhagic colitis and hemolytic-uremic syndrome. While early studies suggested that antibiotic treatment may worsen the pathology of an enterohemorrhagic Escherichia coli (EHEC) infection, recent work has shown that certain non-Shiga toxin-inducing antibiotics avert disease progression. Unfortunately, both intestinal bacterial infections and antibiotic treatment are associated with dysbiosis. This can alleviate colonization resistance, facilitate secondary infections, and potentially lead to more severe illness. To address the consequences in the context of an EHEC infection, we used the established mouse infection model organism Citrobacter rodentium ϕstx2_dact and monitored changes in fecal microbiota composition during infection and antibiotic treatment. C. rodentium ϕstx2_dact infection resulted in minor changes compared to antibiotic treatment. The infection caused clear alterations in the microbial community, leading mainly to a reduction of Muribaculaceae and a transient increase in Enterobacteriaceae distinct from Citrobacter. Antibiotic treatments of the infection resulted in marked and distinct variations in microbiota composition, diversity, and dispersion. Enrofloxacin and trimethoprim/sulfamethoxazole, which did not prevent Shiga toxin-mediated organ damage, had the least disruptive effects on the intestinal microbiota, while kanamycin and tetracycline, which rapidly cleared the infection without causing organ damage, caused a severe reduction in diversity. Kanamycin treatment resulted in the depletion of all but Bacteroidetes genera, whereas tetracycline effects on Clostridia were less severe. Together, these data highlight the need to address the impact of individual antibiotics in the clinical care of life-threatening infections and consider microbiota-regenerating therapies.

IMPORTANCE

Understanding the impact of antibiotic treatment on EHEC infections is crucial for appropriate clinical care. While discouraged by early studies, recent findings suggest certain antibiotics can impede disease progression. Here, we investigated the impact of individual antibiotics on the fecal microbiota in the context of an established EHEC mouse model using C. rodentium ϕstx2_dact. The infection caused significant variations in the microbiota, leading to a transient increase in Enterobacteriaceae distinct from Citrobacter. However, these effects were minor compared to those observed for antibiotic treatments. Indeed, antibiotics that most efficiently cleared the infection also had the most detrimental effect on the fecal microbiota, causing a substantial reduction in microbial diversity. Conversely, antibiotics showing adverse effects or incomplete bacterial clearance had a reduced impact on microbiota composition and diversity. Taken together, our findings emphasize the delicate balance required to weigh the harmful effects of infection and antibiosis in treatment.

INTRODUCTION

Infections with bacterial pathogens are still among the most common causes of death globally (1). With the discovery of penicillin in 1928 (2), mortality rates associated with bacterial infections decreased significantly (3). However, research has also shown that the use of antibiotics increases the development of antibiotic-resistant pathogens (1, 3, 4), can enhance susceptibility to other intestinal pathogens (5, 6), or can increase the risk of complications, as in case of infections with enterohemorrhagic Escherichia coli (EHEC) (7–9).

Infections with enterohaemorrhagic E. coli cause bloody diarrhea and can progress to hemorrhagic colitis and hemolytic-uremic syndrome (HUS) in approximately 10%–15% of cases (10). Children are most susceptible to infection, but adults, especially the elderly, are also affected. Shiga toxin (Stx) is considered the major virulence factor of EHEC (7, 10–12) and is responsible for its increased virulence compared to enteropathogenic Escherichia coli (EPEC) (11, 12). To date, the treatment options are still limited and primarily consist of supportive therapies as it has been suggested that antibiotic treatment may cause a worsening of the disease due to increased production and release of Shiga toxin (8, 9). Several studies have investigated the ability of different antibiotics to induce Shiga toxin production in vitro and in vivo (7–9, 13), but the results obtained were inconclusive.

Due to the high host specificity of the pathogen, no proper mouse model was previously available to study EHEC infections in vivo. While Citrobacter rodentium is a suitable model for EPEC infections (14), it does not encode Shiga toxin. In 2012, a C. rodentium strain was generated, which encodes the Stx2_dact phage (C. rodentium ϕstx2_dact), mimicking EHEC and providing a model to study its pathogenesis in vivo (15). Using this mouse model, we demonstrated that the application of stx-inducing antibiotics resulted in weight loss and kidney damage despite the clearance of infection (16). However, several non-stx-inducing antibiotics cleared the bacterial infection without causing Stx-mediated pathology, suggesting that these antibiotics might be useful for treating EHEC infections (16).

Unfortunately, many antibiotics are known to alter the composition and richness of the microbiota, resulting in dysbiosis, which can be unfavorable for overall health (17, 18). The intestinal microbiota of mammals has a significant impact on metabolic and nutritional activities (19, 20) and host immune responses [e.g., immune cell population, cytokine patterns (21–23), and behavioral patterns (24)]. An intact microbiota provides beneficial biological functions by producing metabolic products such as vitamins and short-chain fatty acids (SCFAs) (25). It is thus not surprising that antibiotic-mediated changes in the microbiota can be associated with intestinal disorders such as inflammatory bowel disease, diarrhea, and colitis, as well as extraintestinal and systemic disorders, including metabolic diseases (diabetes), autoimmune responses (rheumatoid arthritis), allergies, asthma, obesity, and even neoplastic and neurodegenerative diseases (19, 26–28).

Another consequence of antibiotic-triggered dysbiosis is the loss of colonization resistance against pathogens (26). A prominent example is treatment with antibiotics such as clindamycin and vancomycin, which leads to increased and long-lasting susceptibility to Clostridioides difficile and allows expansion and dense colonization of resistant Enterococcus and Klebsiella pneumoniae strains (5, 6). Recently, it has also been shown that colonization of mice with C. rodentium and C. rodentium ϕstx2_dact-mediated pathology varies greatly, depending on the intestinal microbiota composition and the production of SCFAs (29).

Although several studies have investigated the changes in the murine gastrointestinal microbiota composition in response to antibiotic treatment (30–37), there is little information on the individual influence of different classes of antibiotics on the microbiota composition in a murine infection model. Additionally, there are no studies that assess the impact of infection on the fecal microbiota in the presence and absence of antibiotics with Shiga toxin-producing C. rodentium. For this reason, we studied the impact of C. rodentium ϕstx2_dact infection and antibiotic treatment on the fecal microbiota composition as a measure of intestinal microbiota disruption. We found that C. rodentium ϕstx2_dact infection caused significant alterations in the composition of the murine fecal microbiota. These changes were, however, minor in magnitude and direction compared to the effects of antibiotics, which resulted in tremendous and highly diverse antibiotic-specific changes in microbiota structure and diversity. Knowledge of the destructive effect on potentially beneficial commensals triggered by a specific class of antibiotics may be helpful for clinical applications, such as treating enteric infections.

RESULTS

Antibiotics are commonly used to treat bacterial infections. However, as some antibiotics are known to induce the expression and secretion of Shiga toxin, they are generally not suggested as a treatment strategy for EHEC infections (7, 8). In an earlier study, we systematically analyzed the effect of antibiotics from different classes on Shiga toxin-mediated disease using C. rodentium ϕstx2_dact. In this study, mice were orally infected with C. rodentium ϕstx2_dact. Starting from day 4 post-infection, mice were daily treated with five antibiotics of different classes [enrofloxacin (Enf), kanamycin (Kan), rifampicin (Rif), tetracycline (Tet), and trimethoprim/sulfamethoxazole (T/S)]. Throughout the study, fecal samples were collected to allow analysis of the microbiota composition on days 0 (before infection), 4, 6, and 12 post-infection (Fig. 1). These time points were chosen as C. rodentium initiates colonization of the colon on day 4, reaches a maximal uniform distribution along the entire colonic mucosa on day 6 (D6), and starts to decline on day 12 (D12) post-infection (38). In total, we analyzed 410 samples of 125 mice belonging to seven different treatment groups [uninfected, infected untreated, and infected, antibiotic-treated mice (Enf, Kan, Rif, Tet, T/S)] at the four time points (Fig. 1) by 16S rRNA gene sequencing assessing the effects of C. rodentium ϕstx2_dact infection as well as short- and long-term antibiotic treatment in C57BL/6Rj mice (Table S1). In total, 7,654,419 bacterial 16S rDNA sequence counts were obtained with a mean of 18,669 ± 8,447 counts per sample.

Fig 1.

Experimental setup. For each experiment, purchased mice were separated into groups of five mice per cage. Mice were infected with C. rodentium ϕstx2_dact by feeding on day 0 (D0), and weight was monitored daily (16). A group of control mice was kept uninfected. Prior to infection and on days 4, 6, and 12 post-infection, fecal samples were collected from which genomic DNA was isolated. From these samples, the V1/V2 regions of the 16S rDNA were amplified and sequenced for microbiota analyses. Three separate experiments were conducted, yielding a total of 15 mice per treatment group. This figure was created with Biorender.com.

Elimination of batch differences by mixing treatment groups

Differences in the intestinal microbiota composition of mice have been described and may depend on age, gender, genetic background, housing conditions, and others (30). We performed five experiments using independently purchased mouse batches (batches 1–5). Each experiment consisted of different treatment groups to investigate the effect of C. rodentium ϕstx2_dact infection and antibiotics on the gut microbiota independently of the mouse batches (Fig. 1). We first assessed whether the mouse batches, although purchased from the same vendor and barrier, differed. For this purpose, we compared the fecal microbiota of all mice on day 0 (D0). Permutational multivariate analysis of variance (PERMANOVA) revealed that there were significant differences (P < 0.001) between the different mouse batches at all taxonomic levels from sequence type up to phylum (Table S2). These differences are also represented in the non-metric multidimensional scaling (nMDS) plot (Fig. S1A). PERMANOVA comparisons between the different experiments showed that all batches were significantly different at the sequence-type level (Table S2). Additionally, there were significant differences in sequence-type richness (ST), evenness (J), and diversity (Fig. S1B).

As every batch of mice was separated into groups of five mice, which were then infected and later divided into different treatment groups (by cage; see Fig. 1), we then determined whether the observed differences were also significant when the treatment groups were compared, or whether the fact that the batches were all separated into the different treatment groups was enough to eliminate the observed variability. Here, PERMANOVA revealed no differences in the microbiota composition at any taxonomic level (sequence type to phylum, Table S2). This can also be seen in the nMDS plot (Fig. 2A). Also, there were no significant differences in sequence-type richness, evenness, or diversity (Fig. 2B). Hence, although there are significant differences in the purchased mouse batches before infection, splitting the mice into different treatment groups abrogated these variations.

Fig 2.

Batch mixing eliminates differences in treatment groups prior to infection. (A) The global bacterial community structure in mouse feces at the start of the experiment was assessed by non-metric multidimensional scaling. The global community structure is based on standardized sequence-type abundance data, and similarities were calculated using the Bray-Curtis similarity algorithm. Each mouse belonged to one of five batches (batch) and was assigned to one treatment group (indicated by different colors). A total of 15 mice were in each treatment group. (B) The diversity of the different treatment groups is indicated by total sequence-type number, Pielou’s evenness (J′), Shannon’s diversity (H′), and Simpson’s diversity (1−λ), respectively, and was analyzed using sequence-type relative abundance data as input. Data are based on an ordinary analysis of variance using Tukey’s test for multiple comparisons. The mean is indicated by +, and the median is denoted by a black line. The box represents the interquartile range. The whiskers extend to the upper adjacent value (largest value = 75th percentile + 1.5 × IQR) and the lower adjacent value (lowest value = 25th percentile − 1.5 × IQR), and the dots represent outliers. There was no statistically significant difference in any of the tested indices between any treatment groups.

Influence of C. rodentium ϕstx2_dact infection on the fecal microbiota

To assess the impact of C. rodentium ϕstx2_dact infection on the fecal microbiota composition over time, samples taken from uninfected and infected untreated mice were compared. Uninfected untreated mice showed no significant changes in the microbiota composition down to the sequence-type level over time (Table S3). This is reflected in the clustering of the samples in the nMDS plot (Fig. S2A) and evidenced by the low distance among centroids (of microbiota structures), which ranged between 10.5% and 16.5% among all sampling times (Table S4). Some genera trended to be differentially distributed, but none were significantly different when corrected for multiple comparisons (Fig. 3; Table S5). Furthermore, there was no change in α-diversity at the sequence-type level over time [ST, Pielou’s J, diversity (Simpson’s index, 1−λ, and Shannon’s index (H); Fig. S2B].

Fig 3.

Heatmap showing genera influenced by infection. The evolutionary history to the left was inferred using the neighbor-joining method and is based on representative nearly full-length 16S rRNA gene sequences for all genera given (Table S6). The evolutionary distances were computed using the p-distance method and are given in units of the number of base differences per site. All ambiguous positions were removed for each sequence pair (pairwise deletion option). Evolutionary analyses were conducted in MEGA version 7. The different phyla observed are indicated by color code. Only genera present in >10% of samples are further analyzed. The scale of the heatmap is indicated to the right and covers four orders of magnitude of mean relative abundance data. Changes over time were assessed by Kruskal-Wallis test with Benjamini-Hochberg corrections for multiple comparisons. Groups of samples were considered significantly different if the adjusted P value was <0.05. Taxa differentially distributed over time were further assessed by Dunn’s post hoc test. A significant change in abundance compared to the previous time point is indicated by a bold arrow if the P value is <0.01 and a thin arrow if the P value is <0.05. The arrow direction indicates an increase or a decrease in abundance.

PERMANOVA revealed significant community structure changes after infection with C. rodentium ϕstx2_dact from sequence type to the family level throughout the experiment and specific, tremendous changes upon infection from days 4 to 6 visible at all taxonomic levels (Table S3). Consequently, while the distance among centroids of microbiota in untreated mice was only 11.4% when D0 and day 4 (D4) were compared, it increased to 28.9% when D0 and D6 were compared (Table S4). Concomitant with community structure changes, a clear increase in multivariate dispersion was observed (from 0.861 and 0.908 on D0 and D4, respectively, to 1.459 and 1.397 on D6 and D12, respectively; Fig. S3). Only slight changes in diversity were observed, and a significant increase in J and diversity (1−λ) was visible only from D4 to D6 (Fig. S2B). The time course of Citrobacter abundance could be followed, which increased from a relative abundance of 2.0% on D4 to 5.0% on day 6 before slightly declining to 3.5% on D12 (Fig. 3), similar to what was observed for colonization levels of C. rodentium ϕstx2_dact (Fig. S4B). In addition to the significant change in Citrobacter abundance, significant changes in the abundance of 17 out of 74 genera or genus-level taxa (23%) were detected (Fig. 3; Table S5). The most prominent changes were observed for Duncaniella and Muribaculum of the Bacteroidetes on D6 post-infection, where the relative abundance dropped to roughly 50%, whereas a further decrease during infection was not apparent. Similarly, Prevotella decreased significantly in abundance during early infection (D6) but not during late infection (D12). These observations contrast with Odoribacter, which increased significantly in abundance during early infection (D6), and Bacteroides, which increased significantly only in the late infection phase (D12). The increase in the abundance of Bacteroides could be further defined to the species level, where out of three species observed in the majority of samples, only Bacteroides uniformis increased significantly (see Fig. S5). Outside the Bacteroidetes phylum, the effect on bacterial genera was minor (Fig. 3; Table S5). Of note, Enterobacteriaceae, related to E. coli but distinct from Citrobacter, increased from a mean relative abundance of <0.01% on D4 to 0.58% on D6. Together, these findings show that the C. rodentium ϕstx2_dact infection causes significant changes in the community structure of the murine intestinal microbiota.

Consequences of antibiotic treatment on microbiota composition

We then investigated variations in the relative abundance of bacteria as a consequence of treatment with different antibiotics on D6 (short-term antibiotic treatment) and D12 (long-term antibiotic treatment) post-infection.

Enrofloxacin treatment was previously shown to clear C. rodentium ϕstx2_dact infection within 2 days (Fig. S4C). Unfortunately, while resolving the infection, the antibiotic-induced Shiga toxin expression and release resulted in severe kidney pathology, weight loss, and death (16). As expected, the abundance of Citrobacter was greatly reduced after treatment onset (from a mean of 3.2% on D4 to 0.01% on D6), and the pathogen was eliminated afterward (Fig. 3; Table S5). The reduction of Citrobacter abundance was accompanied by a significant change in community structure as evidenced by PERMANOVA (Table S3; see also visualization in the nMDS plot Fig. S6A) and also seen in the distance among centroids (Table S4), with a slight decrease in evenness from D6 to D12 (0.815 ± 0.056 on D6 to 0.768 ± 0.062 on D12) but no significant effect on richness and diversity (Fig. 4A). The multivariate dispersion increased tremendously from 0.986 and 1.063 on D0 and D4, respectively, to 1.73 and 1.818 on D6 and D12, respectively (Fig. S3), the highest heterogeneity observed here. Fifty-two of 71 genera (73%) showed a significant change in their abundance over time (Table S5). Forty-eight of these were affected during short treatment (D4 vs D6), but only 5 were affected during long-term treatment (D6 vs D12) with enrofloxacin. Interestingly, the majority of genera of the Bacteroidetes, Proteobacteria, and Actinobacteria phyla decreased considerably in abundance during early treatment (see Fig. 6), with all three Bacteroides species practically eliminated already on D6 (Fig. S5C). In contrast, no clear trend was observed within Firmicutes (Fig. 6). Several Ruminococcaceae (e.g., Harryflintia and Anaerotruncus) as well as Oscillibacter and Dysosmobacter increased in relative abundance after short-term antibiotic treatment (D6) but declined later on, whereas an increase in relative abundance of Clostridiales was observed (Fig. 6; Table S5). This trend was also visible in the genera Lachnospiraceae and specifically in unclassified Lachnospiraceae. They comprised a mean of 13.8% on D4 before enrofloxacin treatment and increased significantly to a mean of 48.9%. These differences were also visible when higher taxa (families to phylum) were analyzed (Fig. S7).

Fig 4.

Bacterial community diversity dependent on antibiotic treatment. Diversity is indicated by total sequence-type number, Shannon’s diversity (H′), Simpson’s diversity (1−λ), and Pielou’s evenness (J′), respectively, and was analyzed using sequence-type relative abundance data as input. Differences in diversity were analyzed using a mixed effects model, and multiple comparisons were corrected using Tukey’s test (A, enrofloxacin; B, kanamycin; C, rifampicin; D, tetracycline; E, trimethoprim/sulfamethoxazole) separately over time. Statistically significant differences are indicated as *P < 0.05, **P < 0.01, ***P < 0.001, or ****P < 0.0001. The mean is indicated by +, and the median is denoted by a black line. The box represents the interquartile range. The whiskers extend to the upper adjacent value (largest value = 75th percentile + 1.5 × IQR) and the lower adjacent value (lowest value = 25th percentile − 1.5 × IQR), and the dots represent outliers. A total of 15 mice were in each treatment group.

Treatment with kanamycin also allowed complete elimination of C. rodentium ϕstx2_dact on day 6 post-infection (Fig. S4D) with low colon pathology and no kidney damage (16). However, kanamycin caused the most dramatic changes in microbiota composition (Table S3; Fig. S6B), with 60 out of 65 genera (92%) showing a significant change (mainly a reduction) in relative abundance over time (Fig. 6; Fig. S7; Table S5). In contrast to enrofloxacin, which had a minor impact on diversity and induced only a slight decrease in evenness, kanamycin significantly influenced taxon richness (from 246 ± 69 on D4 to 75 ± 25 on D6), evenness (from 0.797 ± 0.035 to 0.713 ± 0.025), and diversity (H: from 4.36 ± 0.39 to 3.05 ± 0.23, 1−λ: from 0.961 ± 0.020 to 0.899 ± 0.015) (Fig. 4B). A slight recovery in evenness and diversity was observed during long-term kanamycin treatment (D12) (Fig. 4B). Also, in contrast to enrofloxacin, a decrease rather than an increase in dispersion was observed (Fig. S3). In accordance with our previous data (16) and as observed for enrofloxacin, the abundance of Citrobacter was tremendously reduced after short-term kanamycin treatment and completely abolished on D12 (Fig. 6; Fig. S8). Furthermore, all other proteobacterial genera significantly diminished during short-term antibiotic treatment, and bacterial reads that could be classified to any proteobacterial class were absent on D12. Similarly, both actinobacterial genera (Adlercreutzia and Bifidobacterium) and unclassified Eggerthellaceae were nearly abolished on D6, with only a few reads remaining in some communities (Fig. 6; Table S5). Nearly all Clostridiales genera were also practically eliminated already on D6, and unclassified Ruminococcaceae, unclassified Lachnospiraceae, or unclassified Clostridiales followed the same trend with a significant reduction during short-term antibiotic treatment. An exception was bacteria with similarity in sequence to Clostridium fusiformis of the Lachnospiraceae, which increased in relative abundance and reached a mean of 1.4%. The Erysipelotrichiaceae genera Faecalibaculum and Duboisiella also significantly increased in relative abundance upon kanamycin treatment (Fig. 6; Fig. S8; Table S5). Members of the Bacteroidetes showed a mixed behavior, where specifically Bacteroides increased by more than one order of magnitude in relative abundance during short-term antibiotic treatment (from 2.5% to 31.9%). The most extreme change was observed for Bacteroides acidifaciens, which increased from a mean of 0.5% on D4 to a mean of 24.4% on D6, whereas Bacteroides 11 was only slightly affected (Fig. S5). Also, Parabacteroides increased from below 0.1% before antibiotic treatment to a mean of 5.2% relative abundance on D6 and 7.8% on D12, whereas, e.g., Odoribacter and Muribaculum decreased (Fig. 6; Table S5). These differences in abundance upon treatment were also observed at higher taxonomic levels, where overall, only Bacteroidales showed an increase in relative abundance, whereas the abundance of all other phyla decreased (Fig. S7).

Rifampicin also allowed survival and prevented kidney damage by the infection with C. rodentium ϕstx2_dact, but overall, the colon pathology was somewhat higher, as the pathogen was not fully eliminated (16) (Fig. S4E). Microbiota analysis confirmed this result, and Citrobacter remained as an important member of the microbial community, although at a relatively low abundance of 0.2% after 6 and even 12 days (Fig. 6). Rifampicin treatment was accompanied by a substantial reduction in richness (from 233 ± 52 on day 4 to 69 ± 15 on D6) and diversity (H: from 4.291 ± 0.246 to 3.134 ± 0.408, 1−λ: from 0.962 ± 0.014 to 0.913 ± 0.054) that recovered slightly but significantly over treatment time (richness: 125 ± 27 on D12, H: 3.710 ± 0.287, 1−λ: 0.947 ± 0.020) (Fig. 4C). There was a slight increase in multivariate dispersion concomitant with antibiotic treatment (Fig. S3). PERMANOVA shows that rifampicin treatment resulted in significant changes in microbiota composition over time (Table S3), and the nMDS plot revealed a slight recovery of microbiota composition on D12 (Fig. S6C). This was reflected in the number of genera differentially distributed, where out of 60 genera, 47 (83%) were affected during early antibiotic treatment (D4–D6), and 25 were affected during late antibiotic treatment (D6–D12) (Fig. 6; Table S5) and in the distance among centroids, which reached 58.1% after early antibiotic treatment (D4–D6) but also 44.8% after long-term treatment (D6 to D12) (Table S4). As observed for kanamycin treatment, most Clostridiales genera clearly diminished in relative abundance. Also, unclassified Lachnospiraceae diminished dramatically in relative abundance by two orders from D4 to D6 (Table S5). However, they recovered to pre-treatment levels on D12. Interestingly, recovery was also observed for a variety of Ruminococcaceae and Lachnospiraceae genera (see Fig. 6; Fig. S8). In contrast, Erysipelotrichiaceae (Faecalibaculum and Clostridium XVIII) increased significantly after rifampicin treatment. Such an increase in relative abundance was also evident for various proteobacterial genera. However, during extended treatment, they regained their original relative abundance levels. Enterobacteriaceae related to E. coli and distinct from Citrobacter increased from a mean relative abundance of <0.01% on D4 to 9.2% on D6 and then dropped to 0.32% on D12. The effect of rifampicin on Bacteroidetes was typically negative and resulted in relative depletion of Alistipes, Prevotella, Odoribacter, and Parabacteroides, usually by at least one order of magnitude (Fig. 6; Fig. S8; Table S5). Accordingly, depletion was observed for the whole Bacteroidales class as well as for Clostridiales, whereas Erysipelotrichiales followed, opposing abundance effects (Fig. S7).

Similar to kanamycin, tetracycline was also able to fully eradicate Citrobacter colonization on D6 (Fig. S4F), allowed murine survival, and abolished colon and kidney damage (16). The depletion could be confirmed here by microbiota analysis, where the relative abundance of Citrobacter was 0.004% on D6, with no Citrobacter detectable on D12 (Fig. 6; Table S5). Diversity changes during tetracycline treatment were prominent with a severe decline in richness (268 ± 72 to 125 ± 31), J (0.806 ± 0.023 to 0.758 ± 0.039), and diversity (H: 4.479 ± 0.299 to 3.652 ± 0.362, 1−λ: 0.970 ± 0.014 to 0.945 ± 0.019) from D4 to D6 (Fig. 4D). A clear shift in microbiota composition was observable, and while samples obtained on D0 and D4 clustered together in the nMDS plot, the samples for both D6 and D12 showed distinct, separate clustering (Fig. S6D). This suggested successive shifts in microbiota composition throughout treatment as also indicated in the distance among centroids, which reached 63.4% when D4 and D6 were compared but remained high (39.6%) comparing D6 and D12 (Table S4). These changes were statistically significant, as evidenced by PERMANOVA (Table S3). Interestingly, tetracycline addition did not affect multivariate dispersion (Fig. S3). A total of 52 of 66 genera (79%) were influenced by tetracycline treatment, with 47 being impacted from D4 to D6 and 12 genera from D6 to D12 (Fig. 6; Fig. S8; Table S5). The application of tetracycline resulted in a rapid depletion and elimination of Lactobacillaceae, Actinobacteria, and Proteobacteria (Fig. 6; Fig. S7; Table S5). Also, most Clostridiales genera were negatively affected. However, Ruthenibacterium and Lachnospiraceae of the C. fusiformis cluster showed an increase during early treatment. Bacteroidetes showed a different oscillating behavior. All three Muribaculaceae genera (Duncaniella, Muribaculum, and Paramuribaculum) decreased under tetracycline treatment, with both Muribaculum and Paramuribaculum recovering during extended treatment. Similarly, the relative abundance of Prevotellamassilia, Prevotella, and Odoribacter decreased, and that of Prevotellamassilia and Prevotella returned to higher relative abundance levels during extended treatment. In contrast, Alistipes and Bacteroides showed an extreme initial relative abundance increase during tetracycline treatment (Fig. 6; Fig. S8; Table S5). A detailed analysis of the species level revealed that only Bacteroides 11 and B. acidifaciens contributed to this overall increase of the genus, with B. acidifaciens increasing from 0.6% relative abundance on D4 to 40.2% on D12. This contrasts with the severe depletion observed for B. uniformis (Fig. S5F).

In the case of trimethoprim/sulfamethoxazole, complete elimination of C. rodentium ϕstx2_dact was observed only on D12 but not on D6 post-infection (Fig. S4G), and this was sufficient to reduce but not abolish colon pathology or kidney damage (16). Accordingly, microbiota analysis revealed significant Citrobacter levels on D6 (0.014%) (Fig. 6; Fig. S8; Table S5). Changes in diversity and richness were minor, and only slight changes in evenness were recorded as significant (Fig. 4E). Dispersion increased after antibiotic treatment (from 0.946 and 0.943 on D0 and D4, respectively, to 1.579 and 1.612 on D6 and D12, respectively, Fig. S3). Also, the fecal microbiota composition changed significantly (Fig. 6; Fig. S8; Table S3) but less tremendously compared to, for example, kanamycin and tetracycline. This was also evident in the distance among centroids, which reached only 39.9% between the communities at D4 and D6 (Table S4). Only 31 of 69 genera (45%) were significantly affected in their abundance. Of these, 21 were affected during early treatment, and only 5 were affected during extended treatment. For trimethoprim/sulfamethoxazole, the most prominent effect observed was a decrease in the relative abundance of various Bacteroidetes genera such as Prevotella, Odoribacter, Duncaniella, Muribaculum, and Paramuribaculum. Clostridiales genera were only slightly affected, whereas Lactobacillaceae increased in relative abundance (Fig. 6; Fig. S8; Table S5).

In summary, both short- and long-term antibiotic treatment resulted in significant and global shifts in microbiota composition (Fig. 6; Fig. S8; Table S5), which were much more dramatic compared to those observed during infection (Fig. 3; Fig. S8; Table S4). Furthermore, these changes were highly specific for each tested antibiotic, with kanamycin having the most prominent effect. In contrast, trimethoprim/sulfamethoxazole triggered relatively minor differences, and no substantial overlaps between treatment groups were observed (Fig. 5 and 6).

Fig 5.

Differences in global bacterial community structure in mouse feces upon infection and subsequent antibiotic treatment. The global bacterial community structure was assessed by non-metric multidimensional scaling and is based on standardized sequence-type abundance data. Similarities were calculated using the Bray-Curtis similarity algorithm. All treatment groups except control mice (UI) were infected with C. rodentium ϕstx2_dact on day 0. Treatment groups that received antibiotics from day 4 post-infection are indicated by Enf, Kan, Rif, Tet, or T/S. Treatment group “UT” remained untreated. The labels indicate the day post-infection. A total of 15 mice were included in each treatment group. Enf, enrofloxacin; Kan, kanamycin; Rif, rifampicin; Tet, tetracycline; T/S, trimethoprim/sulfamethoxazole; UI, uninfected; UT, infected untreated.

Fig 6.

Heatmap showing genera influenced by antibiotic treatment post-infection. The evolutionary history to the left was inferred using the neighbor-joining method and is based on representative nearly full-length 16S rRNA gene sequences of representatives for all genera given (Table S6). The evolutionary distances were computed using the p-distance method and are in the units of the number of base differences per site. All ambiguous positions were removed for each sequence pair (pairwise deletion option). Evolutionary analyses were conducted in MEGA version 7. The different phyla observed are indicated by color code. Only genera present in >10% of a given treatment group are further analyzed. The scale of the heatmap is indicated to the right and covers four orders of magnitude of mean relative abundance data. Changes over time were assessed by Kruskal-Wallis test with Benjamini-Hochberg corrections for multiple comparisons. Groups of samples were considered significantly different if the adjusted P value was <0.05. Taxa differentially distributed over time were further assessed by Dunn’s post hoc test. A significant change in abundance compared to the previously indicated time point is indicated by a bold arrow if the P value is <0.01 and a thin arrow if the P value is <0.05. The arrow direction indicates an increase or a decrease in abundance.

DISCUSSION

Several recent studies described the impact of single and, in some cases, combinations of antibiotics on the intestinal microbiota in mice (30, 32–36, 39). However, a detailed comparative analysis addressing the impact of different classes of antibiotics on the gut microbiota during treatments to eliminate enteric bacterial pathogens has not been performed. Here, we describe that infection with C. rodentium ϕstx2_dact used to mimic EHEC infections in mice causes significant shifts in the relative abundance of members of the fecal microbiota. However, the overall effect of the infection alone was minor compared to that triggered by the treatment with antibiotics.

Mice in which the infection remained untreated displayed a significant increase in Citrobacter with a maximum relative abundance of ~5% on day 6. This corresponds to the relative abundance previously reported for the intestinal lumen (40). However, the mucosal relative abundance appears to reach higher values (38, 41). While no alterations in microbial diversity upon C. rodentium infection were reported in the literature (40), minor variations in microbiota composition were detected (38, 40–43). Lupp et al. indicated a bloom of Enterobacteriaceae upon infection, yet Hoffmann et al. suggested that this bloom at the mucosa was mainly due to an expansion of Citrobacter itself rather than of other family members (41, 42). Also, Hopkins et al. reported an increase in Enterobacteriaceae, which was restricted to the mucosa, however, without analyzing whether this bloom was solely due to Citrobacter (38). We could clearly show that on day 6 post-infection, an increase in Enterobacteriaceae distinct from Citrobacter occurred, probably due to Citrobacter creating a niche for those Enterobacteriaceae.

The most prominent effect of infection observed in our study was a decrease in Bacteroidetes, specifically due to a decrease in relative abundance of Muribaculaceae (Fig. S8), a bacterial family just recently defined and before subsumed into the Porphyromonadaceae (44). One previous report also described a decrease in Porphyromonadaceae and Prevotella, another a decrease in Lactobacillus abundance upon murine infection with wild-type Citrobacter (42, 43), which is similar to the Stx-expressing Citrobacter variant used here. However, a detailed comparison with other reports is difficult as they limited the analysis to higher taxonomic levels (40), analyzed very few animals (38), or the microbiota of the uninfected, naïve mice used in the different studies varied significantly, which is likely to affect the pathogen colonization and pathogen-triggered changes in the microbiota (35). For example, Hoffmann et al. reported a significant increase in Deferribacteriaceae (42), which were absent from the microbiota of mice analyzed here.

In fact, the naïve microbiota of mice used here varied significantly, depending on the mouse batch, even when obtained from the same supplier. While the significance of these community differences was eliminated by mixing different mouse batches to create treatment groups, differences remained at the individual mouse level and may create disparities in colonization resistance mediated by the naïve microbiota (43). As an example, resistance to C. rodentium and EHEC infection has previously been associated with higher diversity and abundance of butyrate-producing bacteria (29, 45) and higher concentrations of SCFAs (29), and administration of butyrate reduced pathogen-mediated intestinal damage (46). Also, enhancement or erosion of the mucus layer by commensals may contribute to the colonization capability (47). Moreover, EHEC and Citrobacter virulence gene expression is controlled by microbiota-derived substances, which will also influence colonization and associated microbiota alterations (47, 48).

Infections with EHEC are commonly not treated with antibiotics due to the fear of antibiotic-induced Shiga toxin production (7–9). However, as the E. coli O104:H4 outbreak in Northern Germany in 2011 has shown, the necessity may arise to treat patients with antibiotics to prevent deadly outcomes. However, the effects of antibiotics from different classes on the induction of stx expression vary greatly. Antibiotics that interfere with DNA replication (e.g., enrofloxacin and ciprofloxacin) induce the bacterial SOS response and tend to induce Shiga toxin synthesis in vitro. In contrast, antibiotics inhibiting protein synthesis (e.g., kanamycin and tetracycline) or blocking bacterial transcription (e.g., rifampicin) consistently showed no stx induction (7, 9, 13, 16, 49) and enhanced survival of EHEC- or C. rodentium ϕstx2_dact-infected mice (16, 50). In particular, tetracycline and kanamycin cleared an infection with C. rodentium ϕstx2_dact without causing kidney damage (16), whereas rifampicin only reduced the Citrobacter load but limited kidney damage (16). Importantly, several patients who developed HUS during the E. coli O104:H4 outbreak were treated with rifaximin, a rifampicin derivative. All patients survived and had fewer seizures than those not treated (51), suggesting that these antibiotics may provide promising options for life-threatening EHEC infections.

The use of antibiotics is well known to cause dysbiosis (52, 53). However, little information was available regarding the influence of individual antibiotics on the fecal microbial community structure in a comparable EHEC infection setting. Here, we showed that mice treated with individual antibiotics of different classes showed large, antibiotic-specific shifts in microbiota composition and varied in their response to long-term antibiotic treatment. We showed here that the administration of kanamycin, tetracycline, or rifampicin as promising treatment options all resulted in significant abundance changes of at least 75% of genera and genus-level taxa observed and caused a significant reduction in diversity. This dysbiosis may trigger adverse effects, including the opening up of niches for infection with or outgrowth of pathobionts such as Clostridioides difficile and Enterococcus faecalis (5, 6), which should be considered for clinical applications. Furthermore, the microbiota also has some propensity to recover upon cessation of treatment (54, 55). While some of our data suggest that this also occurs with extended treatment, future studies should evaluate the ability of the microbiota to recover from the treatment with various antibiotics, as this factor should also play an important role for clinical considerations.

Several recent studies have also investigated the effects of single antibiotics or antibiotic cocktails on the intestinal microbiota of naïve mice. However, some of these studies used very few animals in the different treatment groups (31, 32, 34), such that the significance of identified differences can only poorly be assessed. For example, Sun et al. (31) used only five mice per treatment group and detected no changes in microbiota composition upon enrofloxacin treatment at the phylum level but an increase in Prevotellaceae and Rikenellaceae and decrease in Bacteroidaceae families (31). This contrasts with our report, where a significant impact on nearly all Bacteroidetes genera and all Bacteroidetes families and various other genera could be evidenced with threefold that sample size. Another study used qPCR on a restricted number of taxa to survey the microbiota. However, the results (56) do not correspond to those of other studies, possibly because the method does not reach the accuracy of 16S rDNA amplicon sequencing or metagenomic analyses. In a metagenomic study assessing the effects of tetracycline, Yin et al. observed a significant decrease in the abundance of Firmicutes together with an increase in Bacteroidetes, in accordance with our results (33), but Zhao et al. obtained slightly different results in their study, which, however, was carried out with an extremely small sample size (34). Namasivayam et al. evaluated the effect of anti-tuberculosis therapy but also of rifampicin alone in small treatment groups (32). Most of the observed community changes were due to rifampicin, and Alistipes, Erysipelotrichiaceae, and Parabacteroides trended to be depleted. Mullineaux-Sanders et al. (47) used six animals per treatment group and showed that kanamycin treatment exhibited severe effects and that the bacterial communities were highly dominated by Bacteroidetes genera, similar to the situation observed in our study. Korte et al. also evaluated trimethoprim/sulfamethoxazole, which induced minimal changes in the community composition (35). Here, we also showed that trimethoprim/sulfamethoxazole did not affect the diversity and caused milder alterations of the microbiota overall. However, this antibiotic was less effective, as it only slowly eliminated the pathogen and reduced but did not abolish Stx-mediated kidney damage (16). Overall, our data confirmed that trimethoprim/sulfamethoxazole and enrofloxacin have a less disruptive effect on the microbiota than tetracycline and kanamycin.

Other studies also included mice with a different, rather unusual naïve mouse microbiota, which also hampered a direct comparison of the antibiotic effects. Very recently, Grabowski et al. (30) aimed to analyze the effect of enrofloxacin; however, the control mice exhibited a tremendously high amount of Carnobacteriaceae (~30%) and Pseudomonadaceae (10%) as unusual gut colonizers, which were rapidly depleted, preventing any detailed evaluation of antibiotic effects. Severe and long-term changes in the intestinal microbiota were also observed with a combination of four different antibiotics (ampicillin, vancomycin, metronidazole, and neomycin) (30, 36, 37, 39). Here, an increase in Enterococcus and a decrease in probiotics-related genera such as Lactobacillus were reported as common across individual and mixed antibiotic treatments (37). However, Enterococcus is not a common member of the murine microbiome and was present here at very low abundance in only a small subset of mice.

A limitation of this study is the housing of five animals per cage. As mice are coprophagic, it has been shown that they can share pathogens as well as protective microbiota in this way (57, 58). Although the same overall trends in microbiota composition changes can be observed for all mice in a treatment group irrespective of their cage group, the use of separate cages would be required to fully mitigate these effects.

The comparison with recent studies made evident that reported antibiotic-mediated effects on the murine microbiota differ significantly, depending not only on the antibiotic and the treatment conditions but also considerably on the composition of the naïve microbiota, the sample size used for analysis, and the sensitivity of the applied microbiota analysis method. All these factors contribute to seemingly inconsistent results between studies. Moreover, additional information is required about inhibitory and enhancing effects associated with individual antibiotics, such as concentrations of metabolites controlling microbial growth and host immune responses, as well as the expression of virulence-relevant genes of intestinal pathogens. A more detailed understanding of the effects of antibiotic treatment on different members of the gastrointestinal microbiota will be of great importance to address these issues.

MATERIALS AND METHODS

Animal ethics

The C57BL/6Rj mice (Janvier) were housed under pathogen-free conditions in accordance with Federation of European Laboratory Animal Science Associations (FELASA) recommendations in biosafety level 3 animal facilities of the Helmholtz Centre for Infection Research, Braunschweig, Germany.

Animal infections

Six-week-old female C57BL/6Rj mice purchased from Janvier Labs (Le Genest-Saint-Isle, France) barrier A02 were separated into cages of five mice and infected with 5 × 10⁸ CFU C. rodentium DBS770 (C. rodentium ϕstx2_dact) following the feeding protocol described in Flowers et al. (59) or left uninfected. From 4 days post-infection, drinking water was supplemented with 2% glucose and either of the following antibiotics: enrofloxacin (0.25 mg/mL), kanamycin (2.6 mg/mL), tetracycline (1 mg/mL), rifampicin (1 mg/mL), or trimethoprim/sulfamethoxazole [Trimetotat oral suspension 48% (Livisto)]. Supplemented water was exchanged daily to ensure continuously high levels of antibiotics. Mice were weighed daily. Each treatment group and the uninfected untreated and infected untreated controls included at least 15 mice in total.

Genomic DNA isolation from feces

For collection of fecal pellets from individual mice, animals were separated into boxes for up to 30 min. Samples for gDNA isolation were collected in Lysing matrix D tubes on D0 (before infection) and on D4, D6, and D12 post-infection. Genomic DNA was isolated using the FastDNA Spin Kit and a FastPrep-24 bead beating grinder (MP Biomedicals, Germany) and eluted in 100-µL H₂O_dd. DNA concentrations were determined using a NanoDrop One/One^C spectrophotometer (Thermo Fisher Scientific).

16S rDNA amplification and sequencing

A two-step PCR-approach was used to amplify the V1/V2 variable region of the 16S rRNA gene. PCR with primers 27Fbif and 338R containing part of the sequencing primer sites as short overhangs (given in italics) (ACGACGCTCTTCCGATCTAGRGTTHGATYMTGGCTCAG and GACGTGTGCTCTTCCGATCTTGCTGCCTCCCGTAGGAGT, respectively) was used to enrich for target sequences (20 cycles). A second amplification step of 10 cycles added the two indices and Illumina adapters to amplicons (60). Amplified products were purified, normalized, and pooled using the SequalPrep Normalization Plate (Thermo Fisher Scientific) and sequenced on an Illumina MiSeq (2 × 300 bases, San Diego, USA).

Bioinformatic and statistical analysis

The fastQ files were analyzed with the dada2 package version 1.12.1 in R (61). The quality-trimming and filtering steps were performed using the filterAndTrim function. Forward and reverse reads were trimmed on the 5′-end by 20 and 19 bases, respectively. Reads were truncated to a length of 240 bases, and a maximum of two expected errors per read was permitted. After denoising and paired-end read merging, chimeras were removed. Remaining non-bacterial sequences (eukaryota, mitochondria, and chloroplast) were manually deleted. Overall, 7,654,419 bacterial 16S rDNA sequence counts were obtained with a mean of 18,669 ± 8,447 reads per sample (Table S1). All samples were re-sampled to equal the smallest library size of 6,708 reads using the phyloseq package, returning 3,098 sequence types (62) (Table S1). Sequence types were annotated based on the naïve Bayesian classification with a pseudo-bootstrap threshold of 80% using RDP set18 (63) (Table S1). Sequence variants were then manually analyzed against the RDP database using the Seqmatch function to define the discriminatory power of each sequence type. Species-level annotations were assigned to a sequence variant when only 16S rRNA gene fragments of previously described isolates of a single species were aligned with a maximum of two mismatches with this sequence variant (64). Relative abundances (in percentage) of sequence types, species, genera, families, orders, classes, and phyla were used for downstream analyses. Calculation of diversity indices (species richness ST, Shannon’s diversity index H, Pielou’s evenness J, and Simpson’s diversity index 1−λ) and multivariate analyses were performed using PRIMER version 7.0.11 (PRIMER-E; Plymouth Marine Laboratory, UK), whereas univariate analyses were performed using Prism version 9 (GraphPad Software, Inc.).

Differences in diversity indices between the different mouse batches (Fig. S1B) and the different treatment groups (Fig. 2B) were tested for by ordinary analysis of variance using the Holm-Sidak test for multiple comparisons. Changes in diversity over time were analyzed using a mixed effects model. Tukey’s test was used for multiple comparisons.

The data matrices comprising 3,098 sequence types, 126 genera, or other taxa were used to construct sample-similarity matrices applying the Bray-Curtis algorithm, where samples were ordinated using nMDS with 50 random restarts (65). Significant differences between a priori pre-defined groups of samples were evaluated using PERMANOVA, allowing for type III (partial) sums of squares, fixed effects sum to zero for mixed terms. Monte Carlo P values were generated using unrestricted permutation of raw data (66). Groups of samples were considered significantly different if the P value was <0.05. The abundances of taxa present in the community of at least 10% of the samples were compared by the Kruskal-Wallis test with Benjamini-Hochberg corrections for multiple comparisons (67). Groups of samples were considered significantly different if the adjusted P value was <0.05. Taxa differentially distributed over time were further assessed by Dunn’s post hoc test. The within-group homogeneity was tested by calculating multivariate dispersion indices with PRIMER. Centroids were calculated by PERMANOVA based on sequence-type Bray-Curtis similarity matrices and were used to calculate dissimilarities between treatment groups.

ETHICS APPROVAL

The protocol was approved by the Niedersächsisches Landesamt für Verbraucherschutz und Lebensmittelsicherheit (permission no. 33.19-42502-04-16/2124). Animals were treated with appropriate care, and all efforts were made to minimize suffering.

DATA AVAILABILITY

Demultiplexed raw data for all the amplicon sequencing paired-end data sets are publicly available at the NCBI Sequence Read Archive (SRA) under BioProject accession number PRJNA1011327.

SUPPLEMENTAL MATERIAL

The following material is available online at https://doi.org/10.1128/aac.00057-24.

Supplemental figure and table Legends. aac.00057-24-s0001.docx.

Overview of supplemental material

Supplemental Figures S1 to S8. aac.00057-24-s0002.pdf.

All supplemental figures in one file.

Table S1. aac.00057-24-s0003.xlsb.

Nucleotide sequences of all sequence variants determined using Illumina-based amplicon deep-sequencing, their phylogenetic annotation and sequence count as well as relative abundance data after rarefying across all 410 fecal samples.

Table S2. aac.00057-24-s0004.docx.

Factors influencing global community structures prior to infection as indicated by PERMANOVA.

Table S3. aac.00057-24-s0005.docx.

Influence of infection and antibiotic treatment on the community structure as indicated by PERMANOVA.

Table S4. aac.00057-24-s0006.docx.

Distance among centroids (% Bray-Curtis dissimilarity).

Table S5. aac.00057-24-s0007.xlsx.

Phylogenetic taxa (genera, families, order, classes and phyla) influenced by infection and antibiotic treatment.

Table S6. aac.00057-24-s0008.docx.

Sequences used as representative for genera identified.

ASM does not own the copyrights to Supplemental Material that may be linked to, or accessed through, an article. The authors have granted ASM a non-exclusive, world-wide license to publish the Supplemental Material files. Please contact the corresponding author directly for reuse.