Ethical approval, funding and recruitment

This study was nested within the wider observational study “Investigating Microbial Colonisation and Immune Conditioning in Preterm Babies” [42] and was prospectively incorporated into its design and regulatory approvals. Preterm infants were recruited at Homerton University Hospital (London, UK), between 23+0 and 31+6/40 weeks, within the first 72 h after birth, between January 2016 and March 2017.

Sample collection, storage and transfer

Stool samples were prospectively collected from birth up to the cessation of the subjects’ participation in the study (37/40 weeks CGA, or 12 weeks of age, whichever was sooner). Samples (when available) were taken on a daily basis within four hours of passage and immediately placed in a universal storage container (without preservatives) in a refrigerator at 4 °C. The samples were subsequently transferred to an ultra-low freezer (− 80 °C) generally within 24 h. Samples were later transferred from the clinical site to the laboratory for long-term storage and subsequent analyses; this was performed at intervals using a specialist medical courier service, with maintenance of the cold-chain throughout achieved through transport on dry ice (at − 79 °C). Duration of time in − 80 °C storage ranged from 5 to 20 months.

Sample selection

2229 samples were collected in total from all 71 subjects in this study; 1536 samples were sent for sequencing. As a quality control measure, slots were reserved for positive controls, negative controls and mock communities. Additionally, for other investigations not pertaining to this study, slots were reserved for PMA-untreated samples and Salinibacter ruber spiked samples. Consequently, 1434 samples were selected for analyses.

Samples were prospectively selected from those available on a pragmatic basis, aiming for:

Regularity of sampling (every 2–3 days, where able and available)

The capture of episodes of potential microbiome instability in greater detail—up to daily sampling (i.e. early life, before/during/after courses of antibiotics, feeding/ respiratory support changes, blood transfusions)

The capture of ‘stable’ infants’ microbiome profiles in greater detail (in order to develop a reliable baseline)

Where samples were unavailable/unsuitable for analysis (e.g. small volume samples; samples depleted in pilot studies), an alternative (temporally adjacent) sample was selected.

The microbiome composition of samples was not known prior to selection.

Library preparation

The DNA library was prepared as per the recommended protocol for the DNeasy® PowerSoil® Kit (Qiagen, Netherlands), with the following adaptations:-

A reduction in starting stool mass from 250 to 20 mg, following evidence from the literature, corroborated by our own pilot study, demonstrating improved extraction yields at reduced volumes. [43]

The addition of a manual homogenisation step, in view of viscous samples

The use of propidium monoazide (PMA) following the standard protocol (Biotium, USA). PMA is a photo-reactive, DNA-binding dye, used to prevent downstream amplification of free DNA (e.g. contaminant, non-viable) in PCR [44, 45].

The 16S rRNA gene V4 region was amplified using Fusion primers (v4.SA501-508, v4.SB501-508, v4.SA701-712, v4.SB701-71; Eurofins, Germany). Samples from which no amplified DNA was subsequently demonstrated (using electrophoresis) were removed from further analyses. Normalisation of amplified DNA was undertaken as per the standard protocol for the SequalPrep™ Normalization Plate (96) Kit (Invitrogen, USA). DNA quantitation was carried out as per the standard protocol for Quant-iT™ PicoGreen® dsDNA Reagent (Invitrogen, USA).

Sequencing and processing

Amplicon sequencing was performed by the Barts and the London Genome Centre, using Illumina MiSeq Technology (2 × 250 bp flow-cell for paired-end sequencing with 5% PhiX DNA).

Processing of the sequencing data was conducted within R [46] using the DADA2 pipeline [47], using the standard protocol (i.e. truncations at position 240 for forward reads, 160 for reverse reads; truncation at a base quality score of 2; no ambiguous bases; maximum 2 expected errors; and automated removal of PhiX DNA). Batch processing was employed, based upon the originating sequencing run, due to the potential for run-specific errors and biases. Positive mock community controls were used to ensure sequencing accuracy (ZymoBIOMICS, USA). Amplicon sequence variants (ASVs) identified by the DADA2 algorithm were taxonomically-assigned using a bespoke, curated, site-specific (i.e. preterm gut) database of 16S rRNA gene sequences [48]. The decontam package [49] in R was used to identify contaminant sequences (sequences disproportionately represented in negative controls); these sequences were manually reviewed and discarded if microbiologically-plausible as a contaminant. Samples with < 5000 sequences were discarded, in view of a risk of under-representing true richness and diversity.

Statistics

Comparative statistics between Groups 1 and 2a were assessed by Independent Samples t-test, Mann–Whitney U-test or Chi-Square test/Fisher’s exact test, where appropriate. Diversity was measured using the inverse Simpson index [50]. The central tendency of taxa varying in relative abundance over time were described using the area-under-the curve (AUC), normalised for duration in study [51]. Analyses and descriptive statistics were weighted, where possible, by number of reads/sample and number of samples/subject. Siblings from a multiple (twin) pregnancy were half-weighted (if within the same sub-group), in anticipation of non-independence in microbiome development. In light of the potential for multiple comparison errors from assessment of many taxa, analyses occurred at a family-level (this additionally allowed the inclusion of the greatest proportion of amplicon sequence variants [ASVs] in analyses); and only taxa with a median, normalised AUC > 1% relative abundance were analysed. In anticipation of the longitudinal nature of analyses, where approximately linear trends were identified, individuals’ summary data were represented by the Theil-Sen regression coefficient [52]. Without a precedent of similar longitudinal microbiome studies, a pragmatic decision was made to aim for an 80% power to detect a one standard deviation difference in diversity progression between groups; this would require 16 subjects/group. It was prospectively noted that subject numbers would be restricted by an absence of an open-ended recruitment period and lack of control regarding the allocation of subjects to groups based upon clinical interventions Additional file 1.

Comparisons

In order to describe longitudinal microbiome progression free from obvious external modulators, a group of stable subjects was first identified from within the study population, who were minimally exposed to postnatal antibiotics. They (Group 1) were pragmatically defined as:-

Having had no episodes of late-onset sepsis

Had received only one course of antibiotics, in the immediate postpartum period

Had > 75% of stool samples obtained outside this period of antibiotic administration – this is to ensure that the assessed pattern of microbiome development reflects the ‘natural’ progression as much as possible, without extraneous influence from concurrent antibiotic use.

This group would act as a baseline, against which microbiome development in other groups could be compared.

Two comparator groups were then identified from within the study population: (2a) during and (2b) after courses of antibiotics. Infants in Group 2a, during a period of antibiotic administration, were defined as:-

The antibiotic course must be for an episode of LOS (not EOS)

The antibiotic course must be preceded by at least five consecutive days without antibiotics

Only one episode of antibiotic use was analysed per subject

There must be at least three samples from within the period of antibiotic use

Samples must be within the period 48 h prior to commencing antibiotics (one pre-antibiotic sample only) and 48 h following antibiotic cessation

From within Group 2a, a sub-group, 2b, was identified in whom microbiome development in the period after antibiotic cessation could be studied. This was defined as:-

The period of assessment must occur in an antibiotic-free interval, following an episode of antibiotic use for LOS

At least three samples from this antibiotic-free period must be available for analyses

Not all subjects in Group 2a could be studied as part of Group 2b, because there were too few samples in the period after antibiotics were stopped. No subjects in Groups 2a/2b were in Group 1, by virtue of antibiotic use for LOS. Subjects were allocated to their respective clinical groups following completion of recruitment, and prior to assessment of their microbiome composition.

Parameters of microbiome development were studied in these three groups, with comparison made between those receiving antibiotics for presumed LOS and those with no antibiotic exposure after the immediate post-partum period (i.e. Group 2a vs. Group 1; Group 2b vs. Group 1). Longitudinal diversity progression, and longitudinal progression of Enterobacteriaceae relative abundance were the two parameters which could be seen to follow roughly linear trajectories across the majority of subjects, and are quantitatively described. Other features of microbiome development in the three groups are noted, but were not appropriate for quantitative longitudinal analyses (due to grossly uneven distributions across subjects, or following inconsistent trajectories).

Preliminary analyses demonstrated that longitudinal diversity progression followed an approximately linear trajectory within three groups. Consequently, longitudinal diversity progression was described on an individual-level, as the rate of change of diversity with time. The mean/median of these individual diversity progressions was used as a summary measure to describe the progression within the wider group (i.e. Group 1/2a/2b).

Longitudinal development of Enterobacteriaceae relative abundance was similarly described. From an inspection of individual profiles, levels could be seen to either rise to a peak then fall, or fall from an initial peak. Consequently, for consistency, the rate of Enterobacteriaceae fall from its highest level was described on an individual level, as the percentage change in relative abundance with time. Again, the mean/median of these individual values was used to describe the wider groups.

Comparison was made between the summary measures of progression in Group 1 vs. Group 2a and Group 1 vs. Group 2b, using two-tailed weighted t-tests, or Mann–Whitney U-tests, as appropriate; statistical significance was defined at p < 0.05.