Association between antibiotics and gut microbiome dysbiosis in children: systematic review and meta-analysis

1Research over recent years has emphasized the importance of the gut microbiome, and its association with health and the immune system. On the one hand, methods of enhancing the microbiome have proved effective. For example, probiotics have been used to reduce the incidence of severe necrotizing enterocolitis in preterm neonates as the gut microbiome is insufficiently developed to regulate the intestinal mucosa, and fecal microbial transplant (FMT) is being used successfully to treat patients with allergic colitis or Clostridium difficile infection., On the other hand, damage to the microbiome has been linked with conditions such as asthma, allergy, juvenile idiopathic arthritis,, type 1 diabetes, obesity, celiac disease, mental illness, Crohn’s disease, and impaired neurocognitive outcomes.

2Although the mechanism of association for these diseases has not been fully explored, antibiotics, one of the most commonly prescribed drugs in children in western populations, appear to disrupt the normal maturation of the microbiome and destabilize it, altering basic physiological equilibria., Antibiotics also seem to affect gene expression, protein activity and overall metabolism of the gut microbiota which may directly influence major organ development and immune functioning. Antibiotic exposure has already been shown to alter the gut microbiome in adults and in neonates., This review sought to systematically examine the research into the association between antibiotic exposure and pediatric gut microbiome disruption.

3The literature search identified 4,688 publications. The process of publication selection is described in Figure 1. Twelve studies met the eligibility criteria, were deemed good (nine studies) or fair (3 studies) in quality and were included in the final analysis. Meta-analysis was carried out on four RCTs that shared the Shannon Index as their outcome measure of the impact of antibiotics up to 14 days after administration. Quality assessments of RCTs are presented in Supplementary Data Figure S1 (Cochrane Risk of Bias Tool).A high risk of bias was found in Wei et al.’s trial with respect to blinding of the outcome in the analysis done at 4 years but there was no such risk with respect to the analysis done at 14 days.Quality assessments (Newcastle–Ottawa Scale) of observational cohort studies are presented in Supplementary Data Table S1 and of cross-sectional studies in Supplementary Data Table S2.,

4The main characteristics of the included studies are summarized in Table 1. There were five randomized controlled trials (RCTs), five cohort studies and two cross-sectional studies. All studies detected changes in composition of the microbiome following antibiotic exposure in 3 main outcomes: reduction in microbiome species richness; reduction in species diversity; and change in taxonomic composition (change in a specific phylum, genus or species). The primary aim varied between studies. The age of participants ranged from new-born to 12 years old. Nearly all studies reported the short-term associations between antibiotic exposure (up to 1 month) and microbiome composition; some also reported longer-term outcomes up to 2 years and 4 years.,,,

5Microbiome richness (Table 6) data were available for 3 studies and are shown in Table 2. Microbiota richness in children exposed to antibiotics was statistically significantly reduced compared to that of children not exposed to antibiotics in all three studies.,, Measures of richness included Operational Taxonomic Unit (OTU) count (see Table 2) and a generic measure of ‘observed richness’. The time between exposure and analysis was ≤ 14 days in 2 studies,, and ≤ 6 months in one study. The reduction in richness reported by Wei et al. had resolved by the time of a second analysis (mean of 223 days following exposure). Korpela et al., found that microbiome richness was reduced for up to 1 year following penicillin exposure and for up to 2 years following macrolide exposure. Parker reported that the significant reduction in species richness was driven through depletion of Proteobacteria (mainly the species Akkermansia mucinophilia) which were particularly susceptible to azithromycin. Three other authors also commented on richness but did not report raw data and hence are not included in Table 2.,,

6Microbiome species diversity was reported by 8 studies.,,,,,,, Data were available for 6.,,,,,, The main diversity outcome measure was alpha-diversity (Table 3). Antibiotic use was associated with a reduction in alpha-diversity (measured by Shannon or Simpson/Inverse Simpson indices) in 4 studies.,,, Initial Shannon diversity indices varied substantially by geographical location (approximate index value of ‘3ʹ in studies in China, Denmark, India, and Ireland to approximate index value of ‘15ʹ in Burkina Faso and Niger). We carried out a meta-analysis of 4 RCTs examining the effect of azithromycin on the microbiome measured by the Shannon Index. We found a statistically significant overall reduction in alpha-diversity (mean difference −0.86 (−1.59 to −0.13, p <.001) (Figure 2).

7Beta-diversity was reported in 3 studies and significantly reduced in 2 of those.,, Bokulich examined the impact of exposure to several different classes of antibiotics: cephalosporins, beta-lactams, macrolides, quinolones, and nitrofurantoin. They found that although microbiome alpha-diversity was unchanged following antibiotic exposure, beta-diversity differed significantly between children exposed to antibiotics and those unexposed (UniFrac distance, permutational MANOVA, R2 < 0.01 p <.001). This means that although the individual diversity index did not change (i.e. wide species variety and abundance) there was a significant change in the types of species found. With regards to azithromycin exposure alone, Wei reported associations with reduced alpha and beta diversity. Doan however found that beta diversity was unaffected (i.e., similar types of species in the two groups) following azithromycin exposure. But there was a 43% decrease in Simpson’s community-level gamma diversity (p <.001) which reflected the overall reduction in bacterial diversity of the treatment group compared to the placebo group. Six of eight studies reporting on species diversity found a significant association between antibiotic use and a reduction in species diversity.

8The major phyla reported in all studies were Actinobacteria, Bacteroidetes, Firmicutes, and Proteobacteria. One study reported the phylum Veruccomicrobia. A significant increase or decrease in the abundance of a particular phylum, genus or species was reported in 10 studies. These results are summarized in Table 4.

9The association between antibiotics and the abundance of genus Bifidobacterium (phylum Actinobacteria) was examined in 9 studies (Table 4). In five studies, antibiotics were significantly associated with reduced abundance of Bifidobacteria.,, Both penicillins and macrolides were associated with a decrease in Bifidobacteria although in some studies there was no change. Comparing macrolides with penicillins, Korpela et al. found that exposure to macrolides was associated with a fourfold decrease in Bifidobacteria but that exposure to penicillins was not associated with Bifidobacteria levels. Fouhy et al. found that a combination of ampicillin and gentamicin was associated with reduced Bifidobacteria initially, but that by 8 weeks levels had returned to that of the control group. At species level, Mangin et al. found amoxicillin exposure was associated with complete disappearance of Bifidobacterium adolescentis but that overall concentrations of Bifidobacteria were not altered.

10The association between antibiotics and the abundance of Bacteroidetes phylum (which includes the genus Bacteroides) was examined in seven studies (Table 4). There was a statistically significant change in 4 studies. The 3 studies that reported an increase in Bacteroidetes examined exposure to a combination of antibiotics including cephalosporins and macrolides.,, One study examining only amoxicillin exposure reported a decrease of the species Bacteroides fragilis. In 3 studies there was no change (studies examining amoxicillin, ampicillin/gentamicin and azithromycin).,,

11The association between antibiotics and the abundance of Firmicutes phylum (which includes the genera Lactobacillus and Clostridium) was examined in seven studies (Table 4).,,,,, A statistically significant decrease was seen in 4 studies following antibiotic exposure.,,, Korpela et al. reported that Lactobacillus levels were reduced for up to 12 months following penicillin use and for up to 24 months following macrolide use. The same study found a nearly 3-fold increase in Clostridium within 6 months of exposure to macrolides only (details of specific species not given). Yassour et al reported a 40% decrease in Clostridium spp. belonging to clusters IV and XIVa (inducers of T regulatory immune cells) in children aged 3 who had had antibiotics.

12The abundance of Proteobacteria following antibiotic exposure was examined in six studies (Table 4). In 5 studies there was a statistically significant change in Proteobacteria following exposure to a variety of antibiotics, however the direction of association was not consistent. At phylum level, 4 studies reported an increase in Proteobacteria following exposure to different antibiotics including penicillins, cephalosporins and macrolides.,,, One study reported a decrease in Proteobacteria following azithromycin exposure only. At species level, a statistically significant increase in E.coli was reported following amoxicillin exposure in children aged 1–2 years, but a statistically significant decrease in E.coli was reported following azithromycin exposure in children aged 6–11 months.

13The association between azithromycin and a reduction in the abundance of phylum Verrucomicrobia was examined in one study (Table 4). This phylum has relatively few species described. Parker et al. examined the association between azithromycin and the species Akkermansia mucinophila which completely disappeared with azithromycin use (p <.003).

14As far as we are aware this is the first systematic review to synthesize the evidence of the association between antibiotic exposure and changes in the microbiome specifically in children. We found evidence of microbiome disruption characterized by changes in richness, diversity, and taxonomic composition. We cannot be sure of the duration of these changes from the data presented as most studies only presented short-term data. The studies were heterogeneous, with variation between studies in participant age, setting, duration of antibiotic exposure, type of antibiotic given, mode of delivery, outcome measures and time between exposure and analysis. These factors may influence the association between antibiotic use and microbiome composition. Evidence of change in a wide range of microbiome characteristics associated with antibiotic exposure requires further investigation and explanation.

15We found evidence that antibiotic exposure was associated with a reduction in both richness and diversity. In particular azithromycin exposure reduced microbiome alpha-diversity by a mean reduction in Shannon index of 0.86. The studies looked at a variety of antibiotics covering narrow to broad-spectrum antibiotics, with macrolides and penicillins representing the antibiotics most commonly studied. Although no specific change in richness or diversity emerged according to antibiotic class, we found evidence that macrolides were associated with more changes in the microbiome than penicillins and with effects that persisted for longer.,

16We also found evidence that antibiotic use was associated with a reduced number of gut bacteria thought to be beneficial. Bifidobacteria (phylum Actinobacteria) and Lactobacilli (phylum Firmicutes) are producers of short-chain fatty acids which have positive effects on mammalian energy metabolism and form the basis of probiotic supplements. The majority of studies, however, did not report changes in these genera at species level which limits our appreciation of the changes in specific species associated with antibiotics. We also found evidence that changes in other beneficial bacteria were associated with antibiotic use. One study reported a decrease in Clostridium clusters IV and XIVa which are inducers of T regulatory immune cells which have a role in regulating or suppressing other cells in the immune system. A second found that Azithromycin was statistically significantly associated with reduced numbers of Akkermansia Mucinophilia. This species has previously been recognized as having anti-inflammatory and immunostimulant properties, and improving intestinal barrier function, endotoxinaemia and insulin sensitivity.

17We found evidence that antibiotics were associated with a rise in Bacteroidetes and Proteobacteria following antibiotic exposure. These phyla include species which have been implicated in serious infection. Although Bacteroides spp. may provide some level of protection from invasive pathogens as a gut commensal, Bacteroides have also been associated with bloodstream infections and abscess formation. However, it cannot be assumed that higher levels of Bacteroides in the gut are the source of these infections. E.coli (Proteobacteria) is a common cause of urinary tract infections and sepsis and a major source of antimicrobial resistance.

18Our review highlights important findings regarding the relationship between antibiotic exposure and microbiome disruption in children. A strength of our study is that we only included studies with named antibiotics which included specific details of antibiotic administration, rather than exposure to ‘antibiotics’ in general. However, several studies included more than one named antibiotic, so in these cases it was not possible to associate a particular change with a specific antibiotic or class. In the majority of studies, the indication for antibiotic use was infection. In one study there was no clinical indication for antibiotic use but associations with changes in the microbiome were still present. This supports the independent association between antibiotic exposure and microbiome disruption, although further studies of this relationship are required.0

19The use of different outcome measures limited our ability to make comparisons between studies. Although the primary outcomes reported in the RCTs were similar, the applicability of the meta-analysis result may be limited by variation in initial Shannon index scores which in turn might reflect microbiome diversity by geographical location. We could find no evidence of agreement in the literature on the definition of a normal Shannon Index. This substantial difference in variation by geographical location does not seem to have been highlighted in the literature previously and may be worth further investigation. Outcomes in observational studies covered a number of indices of richness, diversity, and taxonomical changes which precluded meta-analysis of all studies. This variation is likely to reflect a lack of consensus among researchers about the most suitable outcome measures in addition to the complexity of the microbiome itself.

20The majority of the studies included in the review focussed on microbiome changes over a short time following antibiotic prescription, i.e. less than 1 month. There was limited evidence therefore of the duration of the changes following exposure. Studies that examined effects over time, found that microbiome disruption lasted between 1 and 2 years,, depending on the antibiotic studied. In this interval some children will receive a further course of antibiotics potentially disrupting microbiome recovery., Further studies are necessary to determine the duration of microbiome disruption.

21A systematic review of antibiotic prescribing in neonates (up to 44 weeks gestational age) looked at the effects of antibiotics on the neonatal microbiome and similarly found that antibiotic exposure was associated with reduced gut microbial diversity and reduced colonization rates of protective commensal bacteria, although the quality of evidence was low. A study looking at the gut microbiota of adults also found that antibiotic exposure was associated with a decrease in beneficial bacteria such as Bifidobacterium and butyrate producers and an increase in Enterobactericae (phylum Proteobacteria). The majority of the changes lasted for approximately 45 days, but the microbiome had not fully recovered by 180 days. Studies in mice support the findings of more reduced diversity following macrolide exposure compared to amoxicillin exposure. Cumulative effects on the microbiome of multiple antibiotic courses, delayed microbiome maturation following antibiotics and fewer changes associated with narrow-spectrum antibiotics have all been observed.,

22In conclusion this review has gathered compelling evidence that antibiotic exposure in children is associated with a reduction in richness and/or diversity, and a change in the balance of species in the microbiome with reductions in the numbers of commensal bacteria thought to be beneficial. Studies that looked at the impact on the microbiome for more than 1 month were limited but there is evidence that antibiotics are associated with disruption to the microbiome for up to 2 years. Macrolide antibiotics cause immediate and longer term damage. More detailed understanding of the strength and duration of antibiotic-specific associations with microbiome dysbiosis in children is needed. Evidence should be sought of a causal relationship between antibiotic use in children, gut dysbiosis and subsequent risk of local or systemic pathological changes with repeated courses of antibiotics. In the meantime, healthcare practitioners should consider the potential for damage to the gut microbiome when prescribing antibiotics for children.

23Procedures used in this review were consistent with Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines.

24A review protocol was submitted in advance to PROSPERO, a database of systematic review protocols (registration ID: CRD42018094188).

25Our inclusion criteria were: studies of any design-assessing change in the microbiome associated with named antibiotic exposure; participants aged from 0 to <18 years (excluding pre-term babies); assessment of composition and diversity of the microbiome using a genetic analysis technique; comparable reference group or baseline assessment and adequate statistical analysis. Our exclusion criteria were non-original research; studies investigating the impact of antibiotics in labor on neonates; studies investigating exposure to any intervention which was not a named antibiotic; studies assessing the impact of antibiotics on other systemic microflora only, e.g. skin, nasal; conference abstracts where insufficient data were given and where the study authors did not reply to further enquiries; and non-English language articles.

26The literature search was performed in February 2019. The databases searched were MEDLINE, EMBASE and Web of Science. No restrictions were placed on the publication period. Search terms included both text words and MESH terms. The full search strategy can be viewed in Supplementary data Table S3.

27Papers were screened using Covidence software (Melbourne, Australia) to efficiently identify the most relevant and appropriate papers. The first reviewer (LM) conducted the literature search and imported the references. Duplicate articles were removed. Two reviewers (LM and AG) screened titles and abstracts with respect to eligibility criteria. Full-text articles of potentially relevant studies were independently assessed for eligibility by two reviewers (LM and VR). Any disagreements were reviewed by another reviewer (PW) and resolved through discussion.

28Information was extracted from included studies on the study type, purpose, characteristics of study participants (age, co-morbidities), details of the antibiotic exposure (name, route of administration), time between exposure and microbiome analysis, molecular technique used and study outcomes. Molecular techniques used included Fluorescent in-situ hybridization (FISH) and flow cytometry, 16s RNA sequencing, DNA extraction, Next Generation Sequencing (NGS) and whole-genome shotgun sequencing (see Table 5). We excluded papers that did not name the antibiotic as we could not guarantee that the participants had been exposed to antibiotics.

29eta-analysis was performed where studies shared the same outcome and where output data were available to include in the analysis. We performed a meta-analysis of four RCTs including 390 patients looking at the mean difference in Shannon Index before and after antibiotic exposure. Continuous outcomes were analyzed using an inverse variance model with a 95% CI. Values were reported as mean differences. P-values were two-tailed and statistically significant if p <.05. Statistical heterogeneity quantification was performed using the I2 statistic. Degrees of heterogeneity were defined as none (I 0–20%), low (I 25–49%), moderate (I 50–74.9%) and high (I > 75%). When heterogeneity was quantified as low or above, a random-effects model was used. The meta-analysis was performed using review manager (Revman) for MAC (Version 5.3. Copenhagen: The Nordic Cochrane Center. The Cochrane Collaboration, 2014).

30Observational study quality (cohort and cross-sectional studies) was assessed using a modified version of the Newcastle–Ottawa scale., The Newcastle–Ottawa scale is used to assess quality and biases. Points are assigned on a nine-point scale. LM and PW independently assessed quality factors including: i) comparability of exposed and non-exposed groups; ii) evidence of microbiome assessment prior to exposure; iii) record of antibiotic exposure; iv) confounding factors; and v) statistical analysis. RCT quality was assessed using the Cochrane Risk of Bias Tool. LM and PW independently applied the risk of bias assessments to each RCT. Disagreement was resolved through discussion.

31Additional quality features for RCTs included clear description of inclusion/exclusion criteria and of withdrawals/dropouts.

32The primary outcome measure was the change in bacterial composition of the microbiome. This was measured as the changes in microbiome richness, alpha-diversity or taxonomic composition.,, Secondary outcome measures were beta- and gamma-diversity.,, Microbiome richness score measures the total number of species found in a single sample. Microbiome alpha-diversity score measures the number of individual bacteria from each of the bacterial species isolated from a single sample. Beta-diversity examines the differences in species composition between 2 samples.,, Gamma-diversity measures diversity across many samples taking into account the different species found in each sample. With regards to change in taxonomic composition, the four main phyla reported were Actinobacteria, Bacteroidetes, Firmicutes and Proteobacteria. The various different indices used by authors to quantify these measures are summarized in Table 6.