Distinct impact of antibiotics on the gut microbiome and resistome: a longitudinal multicenter cohort study

Abstract

Background: The selection pressure exercised by antibiotic drugs is an important consideration for the wise stewardship of antimicrobial treatment programs. Treatment decisions are currently based on crude assumptions, and there is an urgent need to develop a more quantitative knowledge base that can enable predictions of the impact of individual antibiotics on the human gut microbiome and resistome.

Results: Using shotgun metagenomics, we quantified changes in the gut microbiome in two cohorts of hematological patients receiving prophylactic antibiotics; one cohort was treated with ciprofloxacin in a hospital in Tübingen and the other with cotrimoxazole in a hospital in Cologne. Analyzing this rich longitudinal dataset, we found that gut microbiome diversity was reduced in both treatment cohorts to a similar extent, while effects on the gut resistome differed. We observed a sharp increase in the relative abundance of sulfonamide antibiotic resistance genes (ARGs) by 148.1% per cumulative defined daily dose of cotrimoxazole in the Cologne cohort, but not in the Tübingen cohort treated with ciprofloxacin. Through multivariate modeling, we found that factors such as individual baseline microbiome, resistome, and plasmid diversity; liver/kidney function; and concurrent medication, especially virostatic agents, influence resistome alterations. Strikingly, we observed different effects on the plasmidome in the two treatment groups. There was a substantial increase in the abundance of ARG-carrying plasmids in the cohort treated with cotrimoxazole, but not in the cohort treated with ciprofloxacin, indicating that cotrimoxazole might contribute more efficiently to the spread of resistance.

Conclusions: Our study represents a step forward in developing the capability to predict the effect of individual antimicrobials on the human microbiome and resistome. Our results indicate that to achieve this, integration of the individual baseline microbiome, resistome, and mobilome status as well as additional individual patient factors will be required. Such personalized predictions may in the future increase patient safety and reduce the spread of resistance.

Trial registration: ClinicalTrials.gov, NCT02058888 . Registered February 10 2014.

Keywords: Antibiotic impact prediction; Antimicrobial resistance; Metagenomics study; Plasmid expansion; Resistome analysis.

Conflict of interest statement

MJGTV is a consultant to Alb Fils Kliniken GmbH, Berlin Chemie, MSD/Merck, and Astellas Pharma; has served at the speakers’ bureau of Astellas Pharma, Basilea, Gilead Sciences, Merck/MSD, Organobalance, and Pfizer; received research funding from 3M, Astellas Pharma, DaVolterra, Gilead Sciences, Merck/MSD, Morphochem, Organobalance, and Seres Therapeutics.

The other authors declare that they have no competing interests.

Figures

Fig. 1.

Antibiotic impact on the gut microbiome. Trajectories of richness, Shannon diversity, and Simpson’s evenness before treatment (T0) and at the end of the observation period (T3) are shown on phylum rank (a) and species rank (b) for both antibiotic treatments. Pink data points are measurements at T0, purple data points at T3. Boxplots indicate the distribution of data. The connecting magenta line shows the means at each time point and their development under treatment. The p value is displayed at the top of each box and indicates statistical significant differences between T0 and T3 within each treatment cohort (paired t-test). Under ciprofloxacin treatment, richness and Shannon diversity decrease significantly while Simpson’s evenness remains stable. In contrast, under cotrimoxazole, loss of richness and diversity is less pronounced and only significant on the phylum rank. c Violin plots illustrate the differences in baseline values between those patients with a positive baseline-endpoint comparison (BEC, green color) and those with a negative (orange color). The group size is displayed in the respective colors. Baseline species Shannon diversity was higher in the group of patients that lost diversity under cotrimoxazole, while patients with no decline or even an increase in diversity had a lower baseline diversity. The same was observed for species Simpson’s evenness under ciprofloxacin. d Based on multivariate regression modeling, the average percentage change per defined daily dose (DDD) is illustrated for each treatment cohort. Under both antibiotics, a loss in diversity was observed. However, no statistically significant difference was detected between both antibiotics. If an additional impact of concurrent medication was detected beside antibiotics in the multivariate models, this has been illustrated by different filling pattern. e Mean cumulative dose for antimicrobial agents in DDDs for the ciprofloxacin cohort and the cotrimoxazole cohort at each sampling time point (T0–T3). The colors indicate the drug classes, administered in either the ciprofloxacin or cotrimoxazole cohort (illustrated in brackets). The cumulative dose of ciprofloxacin was higher than the dose of cotrimoxazole. f Mean emergence and disappearance of species under antibiotic treatment in percentage compared to the species count at baseline. Frequent potentially pathogenic species are displayed. The number of patients with an emergence or disappearance of these species is shown in brackets

Fig. 2.

Antibiotic impact on the gut resistome. a Stacked bar chart of summed length-corrected relative abundances (LCRA) of major antimicrobial resistance gene (ARG) classes at baseline (T0) and over the treatment period (T1–T3). The following ARG classes are depicted: aminoglycosides (AGly), beta-lactamases (Bla), fluoroquinolones (Flq), glycopeptides (Gly), macrolide-lincosamide-streptogramin (MLS), nitroimidazoles (Ntmdz), phenicols (Phe), sulfonamides (Sul), tetracyclines (tet), and trimethoprim (Tmt). b Trajectories of antimicrobial resistance genes quantification by LCRA before treatment (T0) and at the end of the observation period (T3) are shown for both antibiotic treatments. Pink data points are measurements at T0, purple data points at T3. Boxplots indicate the distribution of data. The connecting magenta line shows the means at each time point and their development under treatment. The p value is displayed at the top of each box and indicates statistical significant differences between T0 and T3 within each treatment cohort (paired t-test). Trends for LCRA changes are prominent but do not reach statistical significance. c Two-dimensional kernel estimation density of square root transformed LCRA values of sulfonamide and trimethoprim ARG classes in relation to the administered cumulative antibiotic dose in defined daily doses (DDD). ARG LCRA rises significantly with increasing doses of cotrimoxazole, but not under ciprofloxacin. d Based on multivariate regression modeling, the average percentage change of ARG class LCRA per defined daily dose (DDD) is illustrated for each treatment cohort. Bonferroni-corrected statistically significant differences between both antibiotics (LR p < 0.002) are presented by single asterisks. Significant differences in antimicrobial selection pressure were observed for aminoglycoside, CTX-M, glycopeptide, MLS, nitroimidazole, phenicol, sulfonamide, and trimethoprim ARGs. If an additional impact of concurrent medication was detected beside antibiotics in the multivariate models, this has been illustrated by different filling pattern. e Fluoroquinolone resistance-mediating mutation frequencies increase under ciprofloxacin exposure in patient 512 comparing baseline (T0) and endpoint (T3)

Fig. 3.

Antibiotic impact on the gut plasmidome. a Trajectories of total plasmid abundance, plasmid abundance from proteobacteria, plasmid Shannon diversity, and plasmid Simpson’s evenness before treatment (T0) and at the end of the observation period (T3) are shown for both antibiotic treatments. Pink data points are measurements at T0, purple data points at T3. Boxplots indicate the distribution of data. The connecting magenta line shows the means at each time point and their development under treatment. The p value is displayed at the top of each box and indicates statistical significant differences between T0 and T3 within each treatment cohort (paired t-test). Total plasmid abundance, plasmid abundance from Proteobacteria, and plasmid diversity decreased significantly under ciprofloxacin treatment while plasmid evenness remained stable. In contrast, plasmids were not strongly affected by cotrimoxazole. b Based on multivariate regression modeling, the average percentage change of plasmid characteristics per defined daily dose (DDD) is illustrated for each treatment cohort. Bonferroni-corrected statistically significant differences between both antibiotics (LR p < 0.002) are presented by single asterisks. If an additional impact of concurrent medication was detected beside antibiotics in the multivariate models, this has been illustrated by a different filling pattern (checkerboard pattern = virostatic agents, horizontal stripes = antifungal agents, vertical stripes = virostatic and antifungal agents). Trends for plasmid evenness were significantly different, with a slight increase under ciprofloxacin and moderate decrease under cotrimoxazole. c, d The co-occurrence network displays the relationship between ARG-carrying plasmids from certain taxonomic origins and the ARG classes located on these plasmids at each sample collection time point for the ciprofloxacin cohort (c) and the cotrimoxazole cohort (d). The total plasmid-ARG content is expressed by the line width between plasmid origin and ARG class. The bar on the upper right part of each network row displays the scale of the total plasmid-ARG content (range 1–27). The diagrams in the lower right parts illustrate the Proteobacteria plasmid-ARG content for aminoglycoside, sulfonamide, trimethoprim ARGs, and beta-lactamase A enzymes. The y-axis ranges from 1 to 27 and displays the respective plasmid-ARG content. The ARG classes in the diagrams correspond to the colors of the networks and the legend at the bottom of the graph. Plasmids harboring ARGs from Proteobacteria expanded under cotrimoxazole, while ARG-containing plasmids from all origins declined under ciprofloxacin

Fig. 4.

Links between baseline gut microbiome and resistome alteration under antibiotic pressure. a Spearman’s rank correlation matrix revealed a positive correlation between the resistance score (indicating more positive antibiotic resistance gene selection in patients) and baseline microbiome and plasmid diversity. Pink-colored edgings indicate statistically significant correlation coefficients (p ≤ 0.05). b Scatter graphs with detailed illustration of the relation between baseline microbiome and plasmid diversity as well as between resistance score and baseline microbiome and plasmid diversity

Fig. 5.

Independent contributors that shape the gut resistome along with antibiotic treatment. The graph summarizes the concept of additional independent variables that impact the alterations of the gut resistome under antimicrobial selection pressure caused by antibiotic treatment