The premature infant gut microbiome during the first 6 weeks of life differs based on gestational maturity at birth

Background The impact of degree of prematurity at birth on premature infant gut microbiota has not been extensively studied in comparison to term infants in large cohorts. Methods To determine the effect of gestational age at birth and postnatal exposures on gut bacterial colonization in infants, we analyzed 65 stool samples from 17 premature infants in the neonatal intensive care unit, as well as 13 samples from 13 mostly moderate-to-late premature infants and 189 samples from 176 term infants in the New Hampshire Birth Cohort Study. Gut colonization patterns were determined with 16S rDNA microbiome profiling. Results Gut bacterial alpha-diversity differed between premature and term infants at 6 weeks of age, after adjusting for exposures (p=0.027). Alpha-diversity varied between extremely premature (<28 weeks gestation) and very premature infants (≥28 but <32 weeks, p=0.011), as well as between extremely and moderate-to-late premature infants (≥32 and <37 weeks, p=0.004). Newborn antibiotic use among premature infants was associated with lower Bifidobacterium and Bacteroides abundance (p=0.015 and p=0.041). Conclusion Gestational age at birth and early antibiotic exposure have significant effects on the premature infant gut microbiota.

Illumina-utils (https://github.com/meren/illumina-utils) programs merged paired end reads into consensus full-length v4v5 sequences, trimmed the adapters, and filtered for quality. 66% of the nucleotides in the non-overlapping region were required to have a quality score greater than Q30 score and no more than 3 mismatches in the area of overlap are accepted. Chimeric reads were identified by vsearch (3) and removed from the dataset. Samples resulting in fewer than 10,000 reads were not included in this study.
Taxonomic assignment was performed using the GAST algorithm (4) as described previously (5). All new data generated have been deposited into the National Center for Biotechnology Information Sequence Read Archive (https://www.ncbi.nlm.nih.gov/sra) with accession numbers SRP007679 and SRP064159. Additionally, sequences and taxonomy data were deposited in the Marine Biological Laboratory Visualization and Analysis of Microbial Population Structures (https://vamps.mbl.edu) (6).

Data Analysis
Bacterial alpha-diversity was evaluated using Simpson's diversity index (SDI), which is a measure of both the number (richness) and the relative abundance of genera in a sample (7). We evaluated differences in bacterial alpha-diversity and abundance by building linear mixed effects models that accounted for repeated measures, age (day of life at time of sample collection), and exposures. Where feasible, corrected gestational age (gestational age at birth + day of life) was also adjusted for. We considered unadjusted models as well as models adjusted for exposures.
Rarefaction was performed prior to alpha-diversity analysis. We removed taxa that occurred in fewer than 10 samples from the dataset for analyses of bacterial abundance. If groups existed within the dataset (for example, preterm and term infants), each taxon was required to have at least one sample with non-zero counts within each group. If the nlme function in R still produced an error, then taxa for which nlme could not analyse bacterial abundance differences were removed.
To avoid type I errors when evaluating differences or changes in abundance of a large number of bacterial genera, we controlled the false discovery rate by adjusting p-values for multiple comparisons using the R function p.adjust. A permutational multivariate analysis of variance using distance matrices was used to assess differences in bacterial phylogenetic distances (phylogenetic relatedness) between groups.
This was performed using the R function adonis (from the R package "vegan"). Analysis of dispersion between groups found no evidence of presence of significant dispersion difference by permutation testing (Pr(>F)=0.68). Therefore, adonis results were not due to differences in group dispersions. An ANOVA analysis of the difference in dispersion between groups was also not significant (Pr(>F)=0.665), so there was not significant variation between the two groups. Thus, it is unlikely that dispersion heterogeneity confounds our analysis.
To account for the effects of exposures on bacterial alpha-diversity and bacterial phylogenetic relatedness, we included the exposures in linear mixed effects models and when evaluating bacterial phylogenetic relatedness differences. When adjusting for exposures, samples missing data on at least one exposure were removed. We were not able to account for both the effects of exposures and repeated measures specifically when using the R function adonis due to limitations on equation structure; therefore, infant stool samples were considered independent samples when analyzing bacterial phylogenetic relatedness. Only the 6 week timepoint was used in adonis analyses as it had the largest overlap between the premature and term infants. Day of life 0-5 was also considered as a timepoint, but there would only be n=4 premature infant stool samples to compare to, which would be too large a loss of power to be useful.