Introduction

Stokstad et al.1 first found that the administration of Streptomyces aureofaciens led to a twofold weight increase in chickens, and identified that one of its bacteriocins, chlortetracycline, was responsible for this effect. For over 60 years since then, antibiotics, including mainly glycopeptides, tetracycline, macrolides and penicillins, have been used at subtherapeutic levels to promote weight gain in animals. These continue to be widely used in the United States of America, where more than 13 000 tons of active ingredients have been sold in 2009.2 The Animal Health Institute of America (www.ahi.org) has estimated that without the use of growth promoting antibiotics, the USA would require an additional 452 million chickens, 23 million cattle and 12 million pigs to reach the levels of production attained with current practices.3 From the beginning of their use in agriculture in the 1950s, a similar effect on growth was reported in humans,4, 5, 6 but seems to have been neglected. The link recently identified between the gut microbiota, human obesity7 and malnutrition8, 9 has led researchers to reconsider the impact of antibiotics that are associated with obesity particularly when administered in early infancy (Table 1).

Table 1 Studies reporting a significant weight gain associated with antibiotic administration in humans

Glycopeptides, derived from bacteriocins secreted by bacteria of the Actinomycetales order, are antibiotics largely used in human medicine but linked with weight gain both in animals and humans. Indeed, avoparcin, originally isolated from Streptomyces candidus, was associated with weight gain in farm animals,10 whereas vancomycin, which is isolated from Amycolatopsis orientalis, was associated with significant weight gain and acquired obesity in humans11 in addition to increased adiposity in animals.12 In vitro, glycopeptides affect mainly Gram-positive bacteria due to inhibitory activity against peptidoglycan synthesis. However, Pediococcous and Lactobacillus are not susceptible to glycopeptides, with the notable exception of the L. acidophilus group. Even if the susceptibility depends on the bacterial species within the same genus, treatment with oral vancomycin in humans has been associated with a decrease in Clostridium, Enterococcus, Staphylococcus, Bacteroides and Bifidobacterium. However, an overgrowth of Lactobacillus, Pediococcus and Proteobacteria as Klebsiella, Enterobacter or Citrobacter has also been reported (Table 2).

Table 2 Effect of vancomycin and amoxicillin on gut microbiota in the literature

In a previous study, we found an association between vancomycin and acquired obesity in humans, but gut microbiota was not analyzed.11 In this work, we have tested the weight gain effect associated with vancomycin in a new series of patients. In addition, we analyzed the change in the gut microbiota and tested whether bacterial species that were previously linked with body mass index (BMI)13, 14 were predictive of weight gain after vancomycin treatment.

Patients and methods

Patients

All endocarditis patients who were treated by vancomycin or amoxicillin (taken as a control group based on results of our previous study11) in the departments of Cardiology, Hospital de la Timone, Marseille, France were included (Figure 1) between January 2008 and December 2010. Endocarditis was defined using the modified Duke criteria.15 Patients were treated for at least 4 weeks with or without gentamycin. For each case of endocarditis, the decision for surgery was made by a multidisciplinary discussion following our protocol.16 The use of any antibiotic other than amoxicillin, vancomycin or gentamycin for more than 7 days and switching between vancomycin and amoxicillin were exclusion criteria. As all the patients were included after the diagnosis of endocarditis, samples were obtained from the first week after the start of antibiotics. The control group included 42 controls whose stools were analyzed in a previous study.14 Collection of data included age, sex, height, weight and type of antibiotic treatment, but other medications or diet were not collected. According to our published protocols,16 most of patients (>80%) received 6 weeks of treatment. Written consent was obtained from each participant and approved by the ethics committee of the Faculty of Medicine La Timone, Marseille, France under number 08–002.

Figure 1
figure 1

Study flowchart.

Analysis of weight change

A standardized questionnaire was used to collect demographical, clinical and therapeutic data in all patients treated by antibiotics. The baseline weight (1 month before the onset of the disease), weight at 1 year and height were collected for each patient based on clinical records, systematic follow-up consultations or phone calls. The %deltaBMI, a %deltaBMI >10% and an acquired obesity (BMI becoming>30 at 1 year) were assessed comparing baseline BMI and BMI at 1 year. The %deltaBMI was calculated as follows: %deltaBMI=(BMI at 1 year−baseline BMI)/baseline BMI × 100.

Analysis of intestinal flora

Starting in April 2009, all patients with an endocarditis diagnosis were ask to provide a stool sample at diagnosis and as much as possible during the 4–6 weeks of treatment. DNA was isolated from the stool as described in Dridi et al.18 The purified DNA was eluted from samples to a final volume of 100 μl and stored at −80 °C until analysis. Real-time PCR analysis was performed on a Stratagene MX3000 (Agilent, Santa Clara, CA, USA) using the Quantitect PCR mix (Qiagen, Courtaboeuf, France) as previously described,19 which targeted Bacteroidetes, Firmicutes, Lactobacillus and M. smithii. Quantification of all bacteria was performed as previously described.18

A second analysis designed for the identification of gut bacteria predictive of weight gain was performed only for samples obtained during the first week of antibiotics in patients for whom both weight and gut flora data were available. A real-time PCR was performed on a Bio-Rad FLX96 and targeted L. reuteri, L. plantarum, L. rhamnosus, B. animalis and E. coli as previously reported.13

Statistical analysis

Student’s t-test or a Mann–Whitney test was used to compare continuous variables between the two groups according to the distribution assessed by the Kolmogorov–Smirnov test. Proportions were compared using a two-sided Barnard’s exact test.20 The analyses were performed using SPSS v21.0 (IBM, Paris, France).

In the first analysis, the changes of BMI at 1 year were compared between the two antibiotic groups (vancomycin or amoxicillin) for all patients with available information. A logistic regression analysis was performed, using a deltaBMI >10% or acquired obesity as the dependent variable. The regression was adjusted by age, sex, surgery and baseline BMI, and was systematically adjusted by antibiotic treatment (vancomycin or amoxicillin).

In the second analysis, the concentrations of Bacteroidetes, Firmicutes, M. smithii and Lactobacillus were compared between all samples by intervention group (vancomycin or amoxicillin) and controls for all patients with an available stool sample. In a preliminary univariate analysis, all significant comparisons were confirmed using Dunn’s multiple comparison test. To account for the intra-individual correlation within the repeated measures (one individual could have several fecal samples), a linear model with mixed effects and a random intercept was used, adjusted for age, sex and baseline BMI.

Finally, to identify a relationship between the initial gut microbiota and weight gain at 1 year, proportions of endocarditis patients with weight gain were compared between carriers and non-carriers for each bacterium analyzed on samples obtained during the first week of antibiotics. When comparing %deltaBMI at 1 year between carriers and non-carriers, a bilateral Student’s t-test was performed with Welch’s correction after the Kolmogorov–Smirnov test.

Results

Weight change study

Ninety-seven patients were analyzed for weight change at 1 year (Figure 1). The proportion of patients with an increase for more than 10% BMI was higher in the vancomycin group (5/41 (12.2%)) than in the amoxicillin group (1/56 (1.8%), P=0.038). This finding was confirmed using a logistic regression analysis, adjusted for age, sex, surgery and baseline BMI (Tables 3 and 4). It was not significant for an increase in BMI >5% (9/41 in the vancomycin group versus 7/56 in the amoxicillin group, P=0.23). Acquired obesity was observed in four individuals who had all been treated with vancomycin (P=0.01). One of the individuals with acquired obesity was initially lean (BMI 25) and gained 11 kg, whereas the other three have baseline BMI between 27 and 29.

Table 3 Baseline characteristics of patients from the weight change study
Table 4 Logistic regression of a BMI increase over 10% at 1 year according to antibiotics, age, sex, surgery and baseline BMI

Gut microbiota alteration during antibiotic treatment

One-hundred and forty-nine patients were analyzed for gut microbiota alteration during antibiotic treatment (Figure 1). The amount of Firmicutes was significantly higher in the vancomycin group when compared with controls (P=0.007), whereas the amoxicillin group showed no significant difference compared with the control group (P=0.58) (Figure 2). The amount of Bacteroidetes was also increased in the vancomycin group when compared with controls (P<0.0001). Similarly, the amount of Bacteroidetes was higher in the amoxicillin group (P=0.002). Conversely, despite in vitro resistance to the use of antibiotics,21 the amount of M. smithii was significantly decreased in both groups of patients on antibiotics when compared with controls (P=0.01 for the vancomycin group and P=0.01 for the amoxicillin group). In addition, there was no significant difference in the amount of M. smithii between patients receiving vancomycin and amoxicillin (P=0.80). Lactobacillus was significantly increased in the vancomycin group compared with controls (P=0.04). There was no significant difference in the amount of Lactobacillus between the amoxicillin group and the control group. Finally, the total number of bacteria was increased in both groups of patients receiving antibiotics when compared with controls (P<0.0001 and P=0.001 for vancomycin and amoxicillin, respectively). No significant results were obtained in the multivariate analysis.

Figure 2
figure 2

Global modification of the gut microbiota during long-term amoxicillin or vancomycin treatment. *P<0.05, **P<0.005, ***P<0.0005 compared with controls. All the significant comparisons have been confirmed after Dunn’s multiple comparisons test.

Case of a patient with acquired obesity while being initially lean

A 51-year-old woman treated with vancomycin after a valve replacement surgery presented a massive weight gain over the course of a year (height: 1.50 m; weight: 58 kg before endocarditis increasing to 69 kg at 1 year; BMI: 25.8 before the onset of the disease increasing to 30.6, 1 year after the treatment corresponding to an increase of 11 kg, 4.8 kg m−2 and 19% BMI). The analysis of her first stool sample collected promptly after the beginning of the vancomycin treatment was devoid of L. reuteri, L. plantarum, L. rhamnosus, B. animalis and E. coli. L. reuteri was found in high concentrations (4.7 log10 copies DNA per ml) for the first time in her second stool sample, which was collected 68 days after beginning the vancomycin treatment, no E. coli was detected and there was a decrease in Firmicutes (7.9–7.0 log c.f.u. per ml), Bacteroidetes (8.6–7.9 log c.f.u. per ml) and M. smithii (5.8–4.8 log c.f.u. per ml) between the two samples.

Absence of E. coli is predictive of weight gain under vancomycin

In the 21 patients treated with vancomycin for whom a stool sample obtained during the first week of antibiotics was available, the absence of E. coli in the initial gut microbiota was associated with an increase in BMI at 1 year (8/10 with a weight gain versus 3/11 without weight gain, P=0.03). A logistic regression model adjusted for age, sex and surgery identified an association between the absence of E. coli and weight gain at 1 year (odds ratio=10.7; 95% confidence interval 1.39–82.0); P=0.02). This was confirmed using %deltaBMI (mean±standard Error, 4.0±2.1 for patients without E. coli versus −3.4±2.0 for patients with E. coli, P=0.01, Figure 3). This was also true when one outlier (a patient with a BMI increase of 19%) was excluded (n=20, 2.5±1.6 versus −3.4±2.0, P=0.03). We did not find any other significant differences regarding other bacteria (L. plantarum, L. reuteri, L. rhamnosus, B. animalis, Firmicutes, Bacteroidetes, M. smithii and the Lactobacillus genus) or data from the amoxicillin-treated patients. However, the prevalence was extremely low for Lactobacillus and Bifidobacterium, as previously reported.13

Figure 3
figure 3

Weight change at 1 year in patients under vancomycin treatment according to the presence or absence of E. coli in the initial gut microbiota. *P<0.05.

Discussion

In this study, we confirmed that vancomycin was associated with a massive weight gain compared with the weight before the onset of the diseases (endocarditis) in 1/8 human individuals, whereas amoxicillin caused such a massive weight gain in <1 in 50 individuals. We found that the initial gut microbiota was a critical parameter for the identification of individuals more likely to gain weight under vancomycin. Indeed, the presence of E. coli at the beginning of treatment was predictive of weight loss, whereas its absence was predictive of weight gain. In contrast, the only patient with an acquired obesity while being initially lean showed a dramatic increase in L. reuteri. Finally, we confirmed a characteristic profile of the vancomycin gut microbiota with increased levels of Lactobacillus, increased Firmicutes and Bacteroidetes, but decreased levels of M. smithii.

Gut microbiota alterations associated with vancomycin have been previously reported using culture or large-scale molecular studies and included mainly an increase in the Proteobacteria and Lactobacillaceae, and a specific decrease in specific Firmicutes representatives, such as Clostridium, Staphylococcus and Enterococcus. One of the most striking difference between vancomycin and amoxicillin, the latter used in our study as a control, is the increase in Lactobacillus associated with vancomycin (Table 2).

Human obesity, which is promoted by antibiotics when administered in early infancy,22, 23 was associated with a specific profile of the digestive microbiota and a causal role was strongly suggested by the microbiota transplantation experiments.24 Clostridium perfringens,21, 25, 26 C. difficile,21 C. tetani,21 C. botulinum,21 E. coli13 and M. smithii13, 14, 17 have been linked with the absence of obesity. Conversely, Lactobacillus and especially L. reuteri or L. sakei have been linked with obesity.13, 14, 19, 27, 28 A very recent study21 comparing diabetic and nondiabetic women confirms the key role of Lactobacillus in weight regulation in humans as four species, L. fermentum, L. gasseri, L. johnsonii and L. crispatus were positively correlated with BMI, whereas a single species, L. casei was negatively correlated with BMI.21 Consistently, we had previously shown that L. fermentum was associated with weight gain,27 whereas L. casei or L. plantarum were associated with the absence of obesity.14, 27 In one study on mice, L. reuteri L6798 was associated with weight gain, whereas L. reuteri ATCC PTA 4659 was associated with weight loss.29 These results suggest a species and strain specificity in the effect of Lactobacillus probiotics on weight regulation.30 Overall, Lactobacillus provided a biological plausibility for the link between vancomycin and weight gain.

The acquired obesity associated with a dramatic L. reuteri increase, whereas treated by vancomycin in one patient suggest that bacteriocins could be critical in the link between weight and the profile of the gut microbiota. Indeed, Lactobacillus and particularly L. reuteri produce broad-spectrum bacteriocins, which allow for the elimination of various enteropathogens as Listeria monocytogenes, Staphylococcus aureus, Yersinia enterocolitica, Salmonella and E. coli, the latter being associated previously to the absence of obesity.31 In addition, the weight increase in children with Kwashiorkor under treatment with ready-to-use therapeutic food (RUTF) was correlated with an increase of L. reuteri and L. gasseri, and considered to be associated with bacteriocins secretion, resulting in a dramatic gut microbiota alteration.9 Consequently, the effectiveness of antibiotic treatment to combat severe malnutrition8, 32 also supports the role of bacteriocins, as they could increase the ability of the gut microbiota to extract energy from the diet, as found in experimental models.12

Our study includes some limitations. First, all patients had parenteral antibiotic administration. The administration routes have been shown to impact significantly the shift on the gut microbial populations with oral administration being associated with more pronounced alteration of the gut microbiota,33 but intravenous administration was also associated with dramatic gut microbiota changes notably in infants.34 As a confirmation, we found in this study using intravenous route a reproducible alteration previously obtained by oral administration as a Lactobacillus increase under vancomycin. In addition, although large-scale molecular methods became recently accurate at the species level,21 studies targeting bacterial species of interest in the human gut microbiome as in our work remain completely relevant.13, 14 As such, species-specific PCR avoids some biases of large-scale molecular methods such as the primer bias.35

In conclusion, the major weight gain and acquired obesity specifically observed in some patients treated with vancomycin could be related to the selection of resistant bacteria previously associated with obesity, such as an increase in Lactobacillus, and/or to the suppression of bacteria that are associated with the absence of obesity, such as C. perfringens,21, 25, 26 C. difficile,21 C. tetani,21 C. botulinum,21 E. coli13 or M. smithii.13, 14, 17 Bacteriocins, typically secreted by Lactobacillus,31 which are selected by vancomycin, may explain why the effect of an antibiotic on a complex microbial population is much wider than the expected direct ‘in vitro’ effect, by killing populations paradoxically resistant to the antibiotic when used ‘in vivo’. As demonstrated in experimental studies,36 we demonstrated that weight gain effect following antibiotic administration depends both on the initial gastrointestinal microbiota and the antibiotic used. From our results, it seems necessary to inform the patient of the risk of weight gain and obesity during long-term antibiotic treatment, especially when the patient is treated with vancomycin. More generally, this work supports the instrumental role of the human gut microbiota in the pro-obesity effect of antibiotics as recently reported,22, 23 and suggests bacteriocins as a keystone between gut microbiota and obesity.