Strain-specific impacts of probiotics are a significant driver of gut microbiome development in very preterm infants

Abstract

The development of the gut microbiome from birth plays important roles in short- and long-term health, but factors influencing preterm gut microbiome development are poorly understood. In the present study, we use metagenomic sequencing to analyse 1,431 longitudinal stool samples from 123 very preterm infants (<32 weeks' gestation) who did not develop intestinal disease or sepsis over a study period of 10 years. During the study period, one cohort had no probiotic exposure whereas two cohorts were given different probiotic products: Infloran (Bifidobacterium bifidum and Lactobacillus acidophilus) or Labinic (B. bifidum, B. longum subsp. infantis and L. acidophilus). Mothers' own milk, breast milk fortifier, antibiotics and probiotics were significantly associated with the gut microbiome, with probiotics being the most significant factor. Probiotics drove microbiome transition into different preterm gut community types (PGCTs), each enriched in a different Bifidobacterium sp. and significantly associated with increased postnatal age. Functional analyses identified stool metabolites associated with PGCTs and, in preterm-derived organoids, sterile faecal supernatants impacted intestinal, organoid monolayer, gene expression in a PGCT-specific manner. The present study identifies specific influencers of gut microbiome development in very preterm infants, some of which overlap with those impacting term infants. The results highlight the importance of strain-specific differences in probiotic products and their impact on host interactions in the preterm gut.

Fig. 1. Descriptive overview of diet and the preterm gut microbiome in the first 120 d of life (n = 1,431).

a, Proportion of samples where infants were receiving MOM, formula or MOM and formula. b–f, LOESS fits (95% CIs shaded in grey) over time for richness and Shannon diversity (b), aerobic, facultative anaerobic and obligate anaerobic bacteria (c), Gram-positive and Gram-negative bacteria (d), the top four phyla (e) and the top five genera (f).

Fig. 2. Significance and explained variance of 12 clinical co-variates at different timepoints, modelled by ‘adonis’.

Bubbles show the amount of variance (R²) explained by each co-variate at a given timepoint and significant results (FDR < 0.05) are surrounded by a red box. a, Taxonomic profiles at the species level (n = 821). b, Functional metagenomic capacity at the enzyme level, using EC numbers (n = 821).

Fig. 3. Probiotics were the most significant co-variate associated with the microbiome of preterm infants.

a, Transition model showing the progression of samples through each PGCT from DOL 0 to DOL 69, based on probiotic type. The nodes and edges are sized based on the total counts; nodes are coloured according to PGCT and edges by the transition frequency. b, Prevalence of the B. infantis HMO gene clusters among other species. c, Estimated marginal means (95% CIs) representing Shannon diversity for each probiotic type, obtained from the Shannon diversity linear mixed-effects model adjusted for gestational age, birthweight, birth mode, sex, season, antibiotics, day of full feed, MOM, BMF, formula, weight z-scores, DOL and patient ID. The statistical significance shown is after adjustment for multiple comparisons using two-tailed Tukey’s HSD method. d, NMDS plot of taxonomic profiles for all samples (n = 1,431), showing the mean centroid for each probiotic type. e, Prevalence of probiotic species before, during and after probiotic treatment, stratified by probiotic type. Samples from infants who took no probiotic have been subset into three discrete time bins based on the average start and stop days for probiotic treatment (8 DOL and 44 DOL, respectively). The statistical significance shown is within probiotic summary groups (that is, before, during and after) following adjustment for multiple comparisons using Dunnett’s method, whereby samples from infants who took no probiotic were used as the control for each group. f, NMDS plot of EC number profiles for all samples (n = 1,431) showing the mean centroid for each probiotic type.

Fig. 4. Functional implication of PGCTs.

a, NMDS plot of EC number profiles for all samples (n = 1,431) showing the mean centroid for each PGCT. The statistical significance was based on PERMANOVA, with permutations constrained within the patient. b,c, PLS-DA plots of metabolite profiles (n = 50) showing 95% confidence ellipses for each PGCT for stool (b) and serum (c). The statistical significance was based on PERMANOVA. d, NMDS plot of preterm intestinal organoid transcriptome profiles (n = 17; 2–3 per group) showing the mean centroid for each PGCT. CTRL, control. e, Venn diagram showing the number of DEGs compared with control for each PGCT. Zero values were removed for clarity.

Fig. 5. MOM and antibiotics are significantly associated with the preterm gut microbiome.

a,b, Estimated marginal means (95% CIs) representing Shannon diversity for MOM (a) and antibiotics (b), obtained from the Shannon diversity, linear, mixed-effects models adjusted for gestational age, birthweight, birth mode, sex, season, day of full feed, BMF, formula, probiotics, weight z-scores, DOL and patient ID. The statistical significance shown is after adjustment for multiple comparisons using the two-tailed Dunnett’s method, whereby ‘never’ or ‘no’ was used as the control, respectively. c,d, Box plots showing the relative abundance of bifidobacteria and staphylococci in all samples (n = 1.431) across MOM (c) and antibiotic (d) groups. The centre lines denote the median, the box limits denote the IQR and the whiskers extend to the limits. Points outside the whiskers represent outliers. Statistical significance is based on P values and q values obtained from MaAsLin2 analysis.

Extended Data Fig. 1. Sampling overview.

Samples used in the study from birth to day 120. Dashed lines represent the overall mean start and stop day of probiotic treatment.

Extended Data Fig. 2. DMM clustering into PGCTs.

a, Heatmap of all samples (n = 1431) showing the relative abundance of the most dominant species, coloured by phyla, stratified by PGCT. b, Box plots showing the alpha diversity (richness and Shannon diversity) for each PGCT. The centre line denotes the median, the box limits denote the inter-quartile range (IQR) and whiskers extend to the limits. c, LEfSe identifying discriminatory features of each PGCT based on Linear Discriminant Analysis (LDA). Coloured bars denote PGCTs.

Extended Data Fig. 3. Explained variance of 7 clinical co-variates at different timepoints to validate the findings in this study, using a published metagenomic dataset from Olm et al., modelled by ‘adonis’.

Bubbles show the amount of variance (R²) explained by each covariate at a given timepoint. NA values are used when analyses could not be carried out, due to only 1 level of the variable existing in that given timepoint. No results were found to be significant based on taxonomic profiles at the species level.

Extended Data Fig. 4. The individual impact of probiotics and probiotic species on the gut microbiome.

a, Heatmap showing the relative abundance of B. infantis HMO genes and homologs in other species, coloured by species and stratified by probiotic type. b, Locally weighted scatterplot smoothing (LOESS) fit (95% confidence intervals shaded in grey) over time for the top 5 most dominant Bifidobacterium spp. c, Percentage persistence of probiotic species in Infloran® and Labinic™. d, Per-infant per-strain longitudinal abundance of probiotic species in Infloran® and Labinic™ following the cessation of probiotics.

Extended Data Fig. 5. Evidence that monolayers generated from preterm intestinal-derived organoids are differentiated and entirely cover the transwell.

a, Three technical replicates (R1, R2, R3) of alcian blue stained organoid monolayers on transwell inserts. Goblet cells and mucus layer are stained blue indicating the cells have differentiated and the goblet cells are secreting mucus apically. This further demonstrates that the monolayers are polarised and entirely cover the transwell. b, Scanning electron microscopy image at different magnifications showing microvilli resulting from enterocyte differentiation and that cells are contiguous (that is, no holes in monolayer). c, We performed regular visual inspection of the monolayers using light microscopy, and for all monolayers used in this experiment the monolayers showed full confluence across the entirety of the transwell insert.