Swedish cohort.

To confirm findings in the Bangladeshi cohort in an unrelated cohort, samples from the study by Bäckhed et al. (30) of the human gut microbiome in Swedish infants during the first year of life were downloaded from EBI’s Sequence Read Archive under the accession code ERP005989. Briefly, this cohort enrolled 100 mother-infant dyads and used WGS to examine the development of the infant microbiome from birth until age 12 months. For this study, the infant month 4 and month 12 samples were included. After downloading the month 4 and 12 sequencing files, data were processed using AMR++, which includes a normalization step to adjust for differences in read depth, with the same criteria as the Bangladeshi cohort except that the FLASH merge step was skipped due to a longer insert size. As samples were sequenced without selection based on Bifidobacterium levels, correlation between read depth and Bifidobacterium relative abundance was tested using Spearman’s correlation test at both months 4 and 12. Then, regression using the log-transformed total AMR was completed in R 3.4.3 using the glm command and including the number of times that infants that were exposed to antibiotics before age 4 months and the relative abundance of Bifidobacterium in the month 4 samples as predictor variables. Then, to more directly compare results of the two cohorts, samples were classified as high or low Bifidobacterium using the METAXA2 results on the month 4 samples using the same criteria as for the 16S rRNA gene sequencing on the Bangladeshi cohort. The abundances of AMR in the month 4 and month 12 samples were calculated using the AMR++ pipeline as described above.