Background
Mass drug administration (MDA) with azithromycin has been the cornerstone of trachoma control programs since the 1990s and the advent of the SAFE strategy [
1]. There has been considerable research since into the secondary effects of community-wide azithromycin distribution. A study from The Gambia was the first to report ancillary benefits [
2]. They found all-cause illness, fever, diarrhoea and vomiting were reduced for at least 1-month post-treatment compared with topical tetracycline. Similarly, a study from Nepal found reductions in impetigo and diarrhoea up to 10 days post-treatment [
3]. In 2009, Porco et al. reported a 50% reduction in all-cause mortality in children aged 1–9 years in Ethiopia in communities given azithromycin MDA [
4]. This work has been expanded upon in studies across sub-Saharan Africa, demonstrating decreases in diarrhoea, malaria and all infectious mortality [
5‐
11]. Alongside these benefits, there has been evidence of negative effects, primarily the emergence and increasing prevalence of antimicrobial resistance. A study of Aboriginal communities in Australia reported short-term reductions in the prevalence of nasopharyngeal
Streptococcus pneumoniae carriage, but significant increases in macrolide-resistance in identified isolates [
12]. Further studies have supported this increase in macrolide-resistant nasopharyngeal
S. pneumoniae [
3,
13,
14] as well as in faecal
Escherichia coli [
15‐
17], with evidence of macrolide and non-macrolide antimicrobial resistance in the latter.
Better defining the impact of azithromycin MDA on childhood morbidity and mortality was the aim of the MORDOR (Macrolides Oraux pour Réduire les Décès avec un Oeil sur la Résistance) clinical trial, which randomised 1533 communities across three sub-Saharan African countries to four biannual rounds of azithromycin treatment or placebo [
18]. Azithromycin treatment led to an overall reduction in all-cause childhood mortality in targeted communities of 13.5% at 24-month follow-up, although the effect size varied between countries (Malawi; 5.7%, Niger; 18.1%, Tanzania, 3.4%). Secondary analyses found the impact of treatment to be most pronounced in the first 3 months post-treatment and in children aged 1–5 months [
18,
19]. Despite Niger reporting the most significant effect on childhood mortality, there were no significant differences in the causes of mortality between the two study arms in this country [
20]. In both the treatment and placebo arms, malaria (28%), pneumonia (16%) and diarrhoea (14–15%) accounted for the majority of verbal autopsy-confirmed deaths. All-cause mortality was reduced by 9% after treatment in Malawi, while this decrease was not significant, secondary analysis suggested pneumonia and diarrhoea or HIV/AIDS mortality were the drivers of this effect [
21].
Findings from studies nested within MORDOR in Niger supported previous work that reported increases in antimicrobial resistance after azithromycin MDA. The proportion of macrolide-resistant nasopharyngeal
S. pneumoniae isolated was four times greater after treatment [
22]. Doan et al. further explored the impact of treatment on antimicrobial resistance and gut microbiome composition in Nigerien communities through metagenomics [
23,
24]. They found increased prevalence of macrolide resistance after treatment, prevalence was approximately seven times greater after both four and eight bi-annual treatments. Additionally, after eight rounds of treatments, prevalence was also increased for other antimicrobial resistance classes, most prominently β-lactams. Treatment also had a long-term impact on the gut microbiome at 24-month follow-up, reducing diversity, as previously reported, and altering abundance of specific species [
25,
26]. Notably, prevalence of the rarely identified human pathogen
Campylobacter upsaliensis decreased after treatment. However, the majority of affected species have little known role in gut health or pathogenicity. In contrast, findings from a study nested within MORDOR in Malawi which profiled the gut microbiome by 16 S rRNA sequencing, found no change in diversity after treatment and limited impact on individual genera, with only a minor increase in
Prevotella reported [
27].
This study evaluated the impact of azithromycin MDA on the prevalence of gastrointestinal carriage of macrolide-resistant bacteria in communities within the MORDOR Malawi study site, where the observed reduction in childhood mortality after azithromycin treatment was considerably less than in Niger. Additionally, this study investigated changes in the gut microbiome after treatment.
Discussion
This study utilised metagenomic sequencing to examine the impact of azithromycin treatment on carriage of macrolide-resistance bacteria in the gut. Additionally, this study investigated changes in the gut microbiome after treatment. There was significant evidence for increased macrolide resistance after treatment. Gut microbial diversity and global community composition was not impacted by treatment, in agreement with 16S rRNA profiling of a separate longitudinal cohort of children enrolled within MORDOR in Malawi. However, individual species were differentially abundant after treatment, including the putative human enteropathogen Escherichia albertii.
Macrolide resistance increased after four biannual rounds of azithromycin treatment. Individuals with the highest proportion of bacteria carrying macrolide resistance had mostly been treated 6-months prior to sampling as opposed to 12–24-month prior, however there was no significant impact of time-since treatment on macrolide resistance. A study of children aged < 5 years in rural Tanzania surveyed macrolide resistance in faecal
E. coli following community-wide distribution of azithromycin for trachoma control found that post-treatment increase in macrolide resistance waned over time [
16]. The proportion of macrolide-resistant isolates increased sharply in the first month following MDA (16.3% to 61.2%) but had declined significantly by 6 months (31.3%). Studies of perturbation of the gut microbiome following azithromycin treatment consistently find detectable changes within a few days and these can last up to 6 months [
37,
38], however, longer-lasting impact varies between studies [
39,
40]. Future work should ideally include sampling within days of treatment and follow-up beyond 2 years to determine the immediate and enduring impact of azithromycin on faecal carriage and emergence of macrolide resistance.
Prevalence of carriage of at least one macrolide resistant bacterium was higher in this study conducted in Malawian villages compared to those from Niger. At baseline in Niger, the majority of villages had no evidence of macrolide resistance, assessed by metagenomics, in either arm [
23]. By contrast, 58/60 individuals (across both arms) sampled at baseline in Malawi had evidence of macrolide resistance. There is preliminary data from Tanzania suggesting availability of azithromycin at local pharmacies can impact community-level prevalence of antimicrobial resistance [
41]. These Tanzanian communities had received 4 consecutive years of MDA and were revisited 4 years after cessation. At this time, there was a trend towards increased prevalence of macrolide-resistant faecal
E. coli and nasopharyngeal
S. pneumoniae in hamlets where azithromycin was locally available. Despite the differences in baseline prevalence of carriage of macrolide resistant bacteria, the impact of treatment was relatively consistent between Malawi and Niger. Both reported a small reduction in gut microbial diversity after treatment and increased evidence of macrolide resistance, although the latter was determined at the village-level rather than individual-level. However, it is possible that higher baseline levels of resistance to macrolides in Malawi were a factor in the significantly reduced impact of treatment on childhood mortality compared with Niger. This hypothesis could be further evaluated by determining pre-treatment prevalence of macrolide resistance in the Tanzanian communities enrolled in MORDOR, where the lowest impact of treatment was reported [
18] and assessment of antibiotic availability in rural Malawi.
In agreement with findings from Niger, we detected no significant changes in prevalence of resistance to non-macrolide antibiotics after four rounds of treatment [
23]. However, after a further two rounds of treatment in Nigerien communities [
24], resistance to several classes of non-macrolide antibiotics was increased, this was maintained after two additional treatment rounds for β-lactams. Prevalence of resistance to aminoglycosides and trimethoprim was significantly greater after six rounds of treatment in Niger; resistance to these two antibiotic classes was non-significantly increased after four rounds of treatment in our study. The most common explanation for this effect is multi-drug resistant bacteria, driven by shared mechanisms of resistance or genetic linkage, however, carriage of aminoglycoside and trimethoprim resistance was independent of macrolide resistance in this study. It is possible that increases in off-target antibiotic resistance in the gut microbiome of children from studied Malawian communities would reach significance with further rounds of treatment, as reported in Niger.
Five of six bacteria highlighted as common causes of diarrheal disease by the Global Enteric Multicenter Study (GEMS) of infants and young children in developing countries were detectable in this study [
36]. While none were significantly impacted by treatment, combined abundance of these diarrhoeal pathogens was higher after treatment, although this did not reach significance. Focusing on pathogenic
E. coli strains demonstrated a significant increase in abundance of enteropathogenic
E. coli (EPEC) after treatment. Strain-level results should be treated with caution as the majority of reads were not classifiable beyond species-level. A similar increase was observed in
Escherichia albertii, a close relative of
E. coli [
42]. Recent data suggests many gastrointestinal infections classified as
E. coli infections may be the related
E. albertii [
43,
44].
E. albertii possesses many of the virulence factors found in EPEC and multi-drug resistant isolates have been recovered from clinical samples [
45,
46]. Importantly, there was no evidence of macrolide resistance in EPEC or
E. albertii sequences at 24-month follow-up in either arm of the study; it is likely that abundance of this pathogen increased as an effect of treatment rather than it being resistant. For example, it is possible that higher azithromycin susceptibility of other pathogens such as
Campylobacter,
Salmonella and
Shigella may indirectly lead to increased abundance of EPEC and
E. albertii [
47]. Targeted higher-resolution sequencing of
Escherichia to accurately differentiate species and identify strains of
E. coli would be of value.
In this study, the abundance of several
Acinetobacter species increased after treatment, which can be partly explained by their intrinsic resistance to macrolides [
48]. Despite this, the prevalence of the opportunistic pathogens
A. baumanni,
A. johnsonii,
A. pitii and
A. soli is concerning.
A. baumanni infections are often hospital-acquired with the majority of community-acquired infections presenting in individuals with underlying health conditions [
49,
50]. High mortality rates and rapid emergence of antimicrobial resistance suggest this species requires consideration as a serious human pathogen. This was highlighted in this study by evidence of two macrolide-efflux genes (
mphE and
msrE) in an
A. baumanni isolate from a child in the azithromycin treatment arm. Further studies are required to elucidate the impact of these species and whether detection in the gut has any clinical relevance.
The findings presented here are limited to a single geographical zone in the Mangochi district of Malawi, which may limit extrapolation to wider populations. Further to this, participation in faecal sampling was incomplete as approximately 70% of enrolled children provided samples. Given the observed individual-level heterogeneity in this study, it is possible more complete sampling of enrolled children would have led to different outcomes. However, the overlap with results from Niger suggest these effects may be consistent across study sites. Additionally, very few of the individuals sampled herein were aged 1–5 months, the group which saw the greatest reduction in mortality after treatment. It is possible that treatment may have a different impact on the gut microbiome in these younger children.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.