Subjects and sample collection
This study was part of an observational, non-intervention study involving (pre)term infants admitted to the hospital level III NICU or the level II neonatal ward of Isala in Zwolle, the Netherlands. The ethics board from METC Isala Zwolle concluded that this study does not fall under the scope of the Medical Research Involving Human Subjects Act (WMO). Informed consent was obtained from both parents of all individual participants included in the study. For faecal microbiota profiling, 15 late preterm infants (mean ± standard deviation [SD], 35.7 ± 0.9 weeks gestation, 2871 ± 261 g birth weight) were longitudinally sampled during the first six postnatal weeks, resulting in a total of 95 samples. Sampling days for each infant can be found in Table 1. Infants received either no (control), short-term (ST, < 3 days) or long-term (LT, > 5 days) treatment with a combination of amoxicillin and ceftazidime during the first postnatal week. Infants started antibiotic treatment on clinical suspicion of a bacterial infection and, upon negative or positive cultures. Antibiotic administration was respectively stopped (ST) after two to three days or continued (LT) up till a maximum of seven days. Of the LT infants, one was diagnosed with sepsis and three with pneumonia, and, in all cases, the causative pathogen was unknown. Meconium and faecal samples were collected at birth and at postnatal weeks one, two, three, four and six. Samples were stored temporally at − 20 °C until transfer to − 80 °C. Infant clinical characteristics can be found in Table 1. All infants were born vaginally, only received enteral nutrition and did not have clinical signs of food intolerance.
DNA was extracted from faeces by the repeated bead beating plus phenol/chloroform method, as described previously . DNA was quantified using a NanoDrop ND-2000 spectrophotometer (Thermo Fisher Scientific, Wilmington, DE, USA) and by using a Qubit® 2.0 Fluorometer (Life Technologies, Carlsbad, CA, USA), according to manufacturers’ instructions.
Amplification of the V3–V5 regions of the 16S rRNA gene was performed using the Bifidobacterium-optimised 357F and 926Rb primers, as described previously . For each sample, the reverse primer included a unique barcode sequence to allow for multiplexing. Polymerase chain reaction (PCR) and 454 pyrosequencing (GS Junior, Roche) were performed by LifeSequencing S.L. (Valencia, Spain), as described previously . Sequencing data are available in the European Nucleotide Archive (http://www.ebi.ac.uk/ena) under study accession PRJEB19937.
Sequencing data analysis
Pyrosequencing data were analysed using the QIIME software package (v1.8)  applying Acacia , USEARCH , UCLUST  and the SILVA 111 database  for denoising, chimera removal, operational taxonomic unit (OTU) picking and taxonomic classification, respectively. The obtained OTU table was filtered for OTUs with a number of sequences less than 0.005% of the total number of sequences . To account for variation between samples’ total number of reads, rarefaction to 4085 reads per sample was applied.
To identify bacterial taxa that were significantly different in abundance between control, ST and LT infants, the non-parametric Kruskal–Wallis test with Monte Carlo permutation (10,000×) was applied. The Kruskal–Wallis test was done using absolute read counts for each taxonomic group and after the OTU table was filtered for OTUs present in less than 25% of the samples. To compare richness and diversity between samples, the Wilcoxon signed-rank test, the Mann–Whitney U-test and Kruskal–Wallis test were applied for dependent, two groups of independent and more than two groups of independent samples, respectively. To study (dis)similarities in microbiota composition and relate changes in microbiota composition to clinical data, principal component analysis (PCA) and redundancy analysis (RDA) were performed using the Canoco multivariate statistics software v5. For RDA, factors were considered significant when the Bonferroni-corrected p-value was below 0.05. Co-occurrence patterns were determined by Spearman correlation using the taxa that remained after the OTU table was filtered for OTUs present in less than 25% of the samples. Visualisation was done using the Gephi-0.9.1 platform (https://gephi.org) and Adobe Illustrator CS6.
Real-time PCR amplification and detection were performed on a CFX384™ real-time PCR detection system (Bio-Rad). The reaction mixture was composed of 5 μL iQ™ SYBR® Green Supermix, 0.2 μL forward and reverse primers (10 nmol), 1.6 μL nuclease-free water and 3 μL of DNA template (2 ng/μL). Primers used targeted total 16S , Bifidobacterium , Enterococcus  and Enterobacteriaceae . The program for amplification of total 16S, Bifidobacterium and Enterococcus was initial denaturation at 94 °C for 5 min, followed by 40 cycles of denaturation at 94 °C for 20 s, annealing at 60 °C for 20 s and elongation at 72 °C for 50 s, followed by a melt-curve from 60 °C to 95 °C with 0.5 °C steps. The program for amplification of Enterobacteriaceae was initial denaturation at 95 °C for 5 min, followed by 40 cycles of denaturation at 95 °C for 10 s, annealing at 55.8 °C for 20 s and elongation at 72 °C for 20 s, followed by a melt-curve from 60 °C to 95 °C with 0.5 °C steps. Standard curves contained 101–109 16S rRNA copies/μL and were performed in triplicate.
Data were analysed using the CFX Manager™ software (Bio-Rad). Relative abundances of the taxa were determined by dividing the taxa-specific 16S rRNA gene copy number by the total 16S rRNA gene copy number. Quantitative polymerase chain reaction (qPCR) and pyrosequencing data had a Spearman correlation of 0.758, 0.729 and 0.822 for Bifidobacterium, Enterococcus and Enterobacteriaceae, respectively. To identify bacterial taxa that were significantly different in abundance between control, ST and LT infants, the non-parametric Kruskal–Wallis test with Monte Carlo permutation (10,000×) was applied.