Developmental dynamics of the preterm infant gut microbiota and antibiotic resistome

Abstract

Development of the preterm infant gut microbiota is emerging as a critical research priority(1). Since preterm infants almost universally receive early and often extended antibiotic therapy(2), it is important to understand how these interventions alter gut microbiota development(3-6). Analysis of 401 stools from 84 longitudinally sampled preterm infants demonstrates that meropenem, cefotaxime and ticarcillin-clavulanate are associated with significantly reduced species richness. In contrast, vancomycin and gentamicin, the antibiotics most commonly administered to preterm infants, have non-uniform effects on species richness, but these can be predicted with 85% accuracy based on the relative abundance of only two bacterial species and two antibiotic resistance (AR) genes at treatment initiation. To investigate resistome development, we functionally selected resistance to 16 antibiotics from 21 faecal metagenomic expression libraries. Of the 794 AR genes identified, 79% had not previously been classified as AR genes. Combined with deep shotgun sequencing of all stools, we find that multidrug-resistant members of the genera Escherichia, Klebsiella and Enterobacter, genera commonly associated with nosocomial infections, dominate the preterm infant gut microbiota. AR genes that are enriched following specific antibiotic treatments are generally unique to the specific treatment and are highly correlated with the abundance of a single species. The most notable exceptions include ticarcillin-clavulanate and ampicillin, both of which enrich for a large number of overlapping AR genes, and are correlated with Klebsiella pneumoniae. We find that all antibiotic treatments are associated with widespread collateral microbiome impact by enrichment of AR genes that have no known activity against the specific antibiotic driver.

Figure 1. Development of the preterm infant gut microbiota and antibiotic use

a, Six bacterial species are each prevalent (at least one of these species is present in 99.8% of samples) and highly abundant (more than one of these species comprise more than 50% of the community in more than 50% of samples) in the preterm infant gut microbiota. b, Significant developmental trend of abundant species composition over time in control individuals (P < 0.001, Kruskal–Wallis; *P < 0.05, Dunn’s test with correction for multiple comparisons). Significant for all pair-wise comparisons indicated. c, Species richness log fold-change of the preterm infant gut microbiota from directly before to directly after six different antibiotic treatments. Bars represent average species richness log fold-change. Control samples represent the species richness log fold-change over time in individuals with no antibiotic treatment outside the first week of life. Control samples represent similar mean and standard deviation of days between samples as in antibiotic treatment sets. d, Postmenstrual age and human milk significantly increase gut microbiota species richness, while high CRIB (Clinical Risk Index for Babies) score and three specific antibiotics, meropenem, cefotaxime, and ticarcillin–clavulanate, significantly decrease species richness (***P < 0.001, **P < 0.01; *P < 0.05; GLMM with individual as random effect using 401 preterm infant gut samples). Error bars indicate s.e.

Figure 2. Functional metagenomic selections identify novel antibiotic resistance gene clusters in Proteobacterial species in the preterm infant gut

a, Amino acid identity between all functionally selected antibiotic resistance (AR) genes (identified using Resfams) from preterm infant gut resistomes and their top hit in the NCBI protein database (blue, n = 710). Functionally identified AR genes for preterm infant gut resistomes and top hit in AR databases (red, n = 707). Dashed lines represent mean amino acid identity for each group. While we identified known proteins, they have rarely been previously ascribed an AR function. b, The number of unique resistance-conferring fragments originating from MDR species identified from functional metagenomic selections. c, Heatmap representing the number of unique resistance-conferring fragments that originated from each bacterial species to each antibiotic (CH, chloramphenicol; TE, tetracycline; PE, penicillin; CX, cefoxitin; AP, ampicillin; PI, piperacillin; AXCL, amoxicillin-clavulanate; AX, amoxicillin; CZ, ceftazidime; COL, colistin). d, Co-selection of MDR proteins and protein clusters. Circles represent unique AR proteins coloured according to resistance mechanism, where circles representing AR proteins that result in resistance to more than one antibiotic class are twice as large as those representing all other AR proteins. Squares represent antibiotics that were included in functional metagenomic selections coloured on the basis of antibiotic class. Lines connect antibiotic selections to the resistance proteins that conferred resistance to that antibiotic. The heatmap in the lower right corner depicts the percentages of AR genes shared between antibiotic selections.

Figure 3. Species and antibiotic resistance gene enrichment following specific antibiotic treatments

a, Average change in relative abundance of six high-abundance species from directly before to directly after treatment with six different antibiotics. (*P < 0.05, ***P < 0.001; Wilcoxon Rank Sum, Holm–Bonferronni correction for multiple hypotheses; significance tested for averages with absolute value greater than 20%). b, Grey shading represents antibiotic resistance (AR) genes enriched from directly before to directly after treatment at least tenfold in over half of the individuals treated with the specified antibiotic. Pearson correlation coefficient was calculated for all species by all AR genes. Circles represent Pearson correlation for the specified antibiotic by the six high-abundance species across all 401 preterm infant gut microbiota where correlation is greater than 0.2 (P < 0.001, Pearson Correlation, n = 401). Black-outlined circles denote instances where the Pearson correlation coefficient is greater than 0.8. AR genes considered include those identified in functional metagenomic selections. For AR genes identified by functional selection, each antibiotic selection for which the protein was identified is indicated directly to the left of the AR mechanism or class. c, Predictable response of preterm infant gut microbiota species richness to gentamicin and vancomycin combination treatment. Relative abundance of predictive species and AR genes directly before antibiotic treatment with a combination of gentamicin and vancomycin. RPKM, reads per kilobase per million reads.