Short-duration selective decontamination of the digestive tract infection control does not contribute to increased antimicrobial resistance burden in a pilot cluster randomised trial (the ARCTIC Study)

Objective Selective decontamination of the digestive tract (SDD) is a well-studied but hotly contested medical intervention of enhanced infection control. Here, we aim to characterise the changes to the microbiome and antimicrobial resistance (AMR) gene profiles in critically ill children treated with SDD-enhanced infection control compared with conventional infection control. Design We conducted shotgun metagenomic microbiome and resistome analysis on serial oropharyngeal and faecal samples collected from critically ill, mechanically ventilated patients in a pilot multicentre cluster randomised trial of SDD. The microbiome and AMR profiles were compared for longitudinal and intergroup changes. Of consented patients, faecal microbiome baseline samples were obtained in 89 critically ill children. Additionally, samples collected during and after critical illness were collected in 17 children treated with SDD-enhanced infection control and 19 children who received standard care. Results SDD affected the alpha and beta diversity of critically ill children to a greater degree than standard care. At cessation of treatment, the microbiome of SDD patients was dominated by Actinomycetota, specifically Bifidobacterium, at the end of mechanical ventilation. Altered gut microbiota was evident in a subset of SDD-treated children who returned late longitudinal samples compared with children receiving standard care. Clinically relevant AMR gene burden was unaffected by the administration of SDD-enhanced infection control compared with standard care. SDD did not affect the composition of the oral microbiome compared with standard treatment. Conclusion Short interventions of SDD caused a shift in the microbiome but not of the AMR gene pool in critically ill children at the end mechanical ventilation, compared with standard antimicrobial therapy.

The Chao1 index was calculated by using the estimateR function of vegan using the floor function on the calculated RPMR value.Visualization and multiple comparisons were performed as for Shannon's Index.
Beta-distances were calculated against the table of species normalised as RPRM using the vegdist function of vegan with Bray-Curtis distances.To visualise the clustering, we perfomed nMDS using the metaMDS function of vegan using 200 iterations, and vectors 1 and 2 were plotted using ggplot and the geom_point function.The analysis of clustering was performed using Adonis2.To identify separation between groups, Adonis2 was performed for each pair-wise comparison of the above list, and between all timepoints of treatment groups.When comparisons were from a single treatment group (paired or the entire group), Adonis2 was stratified by patient identifier.The resulting p-values from pair-wise and treatmentgroupwise Adonis2 calculations were normalised using FDR.
To visualize the composition of bacterial taxonomy, taxonomic data was aggregated at a Genus level.From the RPMR normalised table, counts were aggregated using the aggregate function of the stats package of R. To identify the 10 most abundant taxa in SDD patients, the median counts of each bacterial genera were calculated BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) using the summarise function of dplyr (Tidyverse suite).Median was chosen as our read-count data is not normally distributed.To identify the 10 most abundant genera in the SC group, the median count for each taxa was calculated, the list of 10 taxa from SDD patients was removed from list of medians, and the next ten most abundant bacteria were selected.All other taxa were aggregated into "Other" to simplify graphing.Standard column graphs were created using ggplot2 and the geom_col function, setting the position to "fill".For continuity and ease of interpretation, the taxa selected from admission samples was used to observe the changes across timepoints.
For the alluvial plots, we used the ggplot2 extension package ggalluvial, and taxonomy from the above boxplots.Alluvial plots were included to highlight changes in the microbiome composition throughout each treatment.
The software package MaAsLin2 v 1.10.0[11] was used to perform further interrogations of the microbiomes.To analyse time dependent change in groups, patient ID was set as a factor and applied as a random effect, sampling number, weight and age were treated as fixed effects and a minimum prevalence of 0.4 was applied.To compare treatment dependent changes, treatment group, weight and age were treated as fixed effects and a minimum prevalence of 0.4 was applied.
Multiple comparisons were corrected using FDR, and all other settings were used as default.Where the difference between more than four taxa was identified by MaAsLin2, the coefficient of change was plotted as horizontal column graph.For interpretation, the absolute value of the coefficient was plotted with the signed direction of change illustrated by column colour.We chose to set both a and the adjusted-a at a rate of 1 in 20.
BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) To compare the top 20 taxa between groups at each timepoint, we calculated the difference for each taxa using a Kruskal-Wallis test, as the KW test is a generalised Mann-Whitney U test.The p-value for each test was adjusted after calculation by applying the FDR correction, with q-values reported.

Analysis of resistance genes:
To focus the analysis of AMR, genes involved in resistance to non-clinically relevant compounds such as heavy metals, detergents, cleaning compounds and tetracyclines were excluded.Statistical analysis and visualization of AMR genes followed the methods outlined previously.
Genes detected with ARIBA were compiled for each patient and measured in Reads per Kilobasegene per Megabase of sequencing (RPKM).
Sequencing and analysis statistics of PICNIC patients is available in Supplementary Table 1 Sequencing and analysis statistics from samples obtained from described by Clark et al [12] are available in Supplementary Table 2.