The antibiotic resistance reservoir of the lung microbiome expands with age in a population of critically ill patients

Abstract

Antimicrobial resistant lower respiratory tract infections are an increasing public health threat and an important cause of global mortality. The lung microbiome can influence susceptibility of respiratory tract infections and represents an important reservoir for exchange of antimicrobial resistance genes. Studies of the gut microbiome have found an association between age and increasing antimicrobial resistance gene burden, however, corollary studies in the lung microbiome remain absent. We performed an observational study of children and adults with acute respiratory failure admitted to the intensive care unit. From tracheal aspirate RNA sequencing data, we evaluated age-related differences in detectable antimicrobial resistance gene expression in the lung microbiome. Using a multivariable logistic regression model, we find that detection of antimicrobial resistance gene expression was significantly higher in adults compared with children after adjusting for demographic and clinical characteristics. This association remained significant after additionally adjusting for lung bacterial microbiome characteristics, and when modeling age as a continuous variable. The proportion of adults expressing beta-lactam, aminoglycoside, and tetracycline antimicrobial resistance genes was higher compared to children. Together, these findings shape our understanding of the lung resistome in critically ill patients across the lifespan, which may have implications for clinical management and global public health.

Fig. 1. Lung resistome of children compared with adults.

A Frequency of children (translucent) and adults (solid) with each antimicrobial resistance gene (ARG), stratified by ARG class (n = 349 patients). Colors indicate the ARG class. B Number of ARGs detected in children and adults by age subgroups (n = 349 patients). Two outliers were omitted for visualization purposes; one 11–18-year-old patient with 18 ARGs detected and another 70–79 year-old patient with 12 ARGs detected. C Number of ARG classes detected in children and adults by age subgroups (n = 349 patients). For Figures B and C, p values were calculated using two-sided Wilcoxon-rank sum test and adjusted for multiple comparisons with False Discovery Rate (FDR) correction. Only the statistically significant p values (p < 0.05) are depicted. Boxplot elements from Figures B and C include a center line (median), box limits (upper and lower quartiles), and whiskers (1.5x interquartile range). Individual data points are shown using overlaid dot plots. D Proportion of patients with ARGs by ARG class, stratified by pediatric and adult cohorts (n = 349 patients). The center dots indicate the proportion and the error bars indicate 95% confidence intervals calculated by the Clopper-Pearson exact binomial method. P values were obtained by two-sided Pearson’s Chi-square test and Fisher’s exact test for comparisons with counts of <5 patients. E Beta diversity of the antimicrobial resistome of children and adults among those with detectable ARGs. The p value was calculated using the Bray-Curtis dissimilarity index and the PERMANOVA test with 1000 permutations. For Figures B to E, the color indicated children (blue) or adults (red). Source data, including all p values, are provided as a Source Data file. Abbreviation: MLS macrolide-lincosamide-streptogramin, TMP-SMX trimethoprim-sulfamethoxazole, NMDS nonmetric multidimensional scaling.

Fig. 2. Association of age with presence of antimicrobial resistance genes (ARGs) in the lung resistome.

Multivariable logistic regression model evaluating the association of (A) binary age and (B) age subgroups with the presence of ARGs, accounting for sex, race/ethnicity, and lower respiratory tract infection (LRTI) status (n = 349 patients). The center squares indicate aOR, and the error bars indicate 95% CIs. The p value was obtained using a two-sided Wald test. Source data are provided as a Source Data file. Abbreviation: aOR adjusted odds ratio, CI confidence intervals, LRTI lower respiratory tract infection, CA-LRTI community-acquired LRTI, HA-LRTI hospital-acquired LRTI.

Fig. 3. Lung bacterial microbiome of children compared with adults.

A Bacterial abundance in the lung microbiome measured in total bacterial alignments to the NCBI NT database per million reads sequenced (NT rpm) in children and adults by age subgroups (n = 349 patients). B Alpha diversity, calculated by the Shannon diversity index, of the bacterial lung microbiome of children and adults by age subgroups (n = 349 patients). Boxplot elements from Figures A and B include a center line (median), box limits (upper and lower quartiles), and whiskers (1.5x interquartile range). Individual data points are shown using overlaid dot plots. C Beta diversity of the bacterial lung microbiome of children and adults. P value calculated based on the Bray-Curtis dissimilarity index and the PERMANOVA test with 1000 permutations. For Figures A to C, the color indicated children (blue) or adults (red). D Statistically significant (p value < 0.05) differential abundant bacterial genera, by log₂ fold change of bacterial counts, detected in children and adults. P values were calculated using a two-sided Wald test adjusted for multiple comparisons. Bar colors indicate whether the species was more abundant in children (blue) or adults (red). E Frequency of the bacterial species detected in ≥5% of children (translucent) and adults (solid) among the differentially abundant bacterial genera. Colors indicate the bacterial genera. For those with multiple species detected per genus, only the most abundant species was included in this analysis. Source data are provided as a Source Data file. Abbreviations: NT rpm nucleotide reads per million reads sequenced, NMDS nonmetric multidimensional scaling.

Fig. 4. Association of age and the bacterial microbiome with presence of antimicrobial resistance genes (ARGs) in the lung resistome.

Multivariable logistic regression model evaluating the association of binary age with the presence of ARGs, accounting for logarithmic total bacterial abundance [log (NT rpm + 0.01)] per sample, bacterial alpha diversity (n = 349 patients). The center squares indicate aOR, and the error bars indicate 95% CIs. The p-value was obtained using a two-sided Wald test. Source data are provided as a Source Data file. Abbreviation: NT rpm nucleotide reads per million, aOR adjusted odds ratio, CI confidence intervals, LRTI lower respiratory tract infection, CA-LRTI community-acquired LRT, HA-LRTI hospital-acquired LRTI.

Fig. 5. Differentially abundant bacterial genera between patients with antimicrobial resistance genes (ARGs) and patients without ARGs in the lung resistome.

Statistically significant (p < 0.05) differentially abundant bacterial genera, by log₂ fold change of bacterial counts, detected in patients with ARGs compared with patients without ARGs. All detected bacterial genera were more prevalent in patients with ARGs compared with patients without ARGs. P values were calculated using a two-sided Wald test adjusted for multiple comparisons. Source data are provided as a Source Data file.