Longitudinal development of the gut microbiome and metabolome in preterm neonates with late onset sepsis and healthy controls

Abstract

Background: Late onset sepsis (LOS) in preterm infants is associated with considerable morbidity and mortality. While studies have implicated gut bacteria in the aetiology of the disease, functional analysis and mechanistic insights are generally lacking. We performed temporal bacterial (n = 613) and metabolomic (n = 63) profiling on extensively sampled stool from 7 infants with LOS and 28 matched healthy (no LOS or NEC) controls.

Results: The bacteria isolated in diagnostic blood culture usually corresponded to the dominant bacterial genera in the gut microbiome. Longitudinal changes were monitored based on preterm gut community types (PGCTs), where control infants had an increased number of PGCTs compared to LOS infants (P = 0.011). PGCT 6, characterised by Bifidobacteria dominance, was only present in control infants. Metabolite profiles differed between LOS and control infants at diagnosis and 7 days later, but not 7 days prior to diagnosis. Bifidobacteria was positively correlated with control metabolites, including raffinose, sucrose, and acetic acid.

Conclusions: Using multi-omic analysis, we show that the gut microbiome is involved in the pathogenesis of LOS. While the causative agent of LOS varies, it is usually abundant in the gut. Bifidobacteria dominance was associated with control infants, and the presence of this organism may directly protect, or act as a marker for protection, against gut epithelial translocation. While the metabolomic data is preliminary, the findings support that gut development and protection in preterm infants is associated with increased in prebiotic oligosaccharides (e.g. raffinose) and the growth of beneficial bacteria (e.g. Bifidobacterium).

Keywords: Gut microbiome; Late onset sepsis; Metabolomics; Preterm infant.

Fig. 1

Area plots showing the temporal development of the microbiome in infants diagnosed with late onset sepsis (LOS). Dashed red lines represent the day of LOS diagnosis with the bacteria isolated from blood culture identified. Dashed black lines represent the start of an antibiotic treatment as per Additional file 2: Table S2.

Fig. 2

Characterisation of the gut microbiome between infants diagnosed with late onset sepsis (LOS) and matched controls. a Transition network analysis showing PGCTs in PreLOS samples compared to matched controls approximated as a Markov chain with subject-independent transition probabilities. Arrow weights reflect the transition probabilities from one sample to the next. Size of circle reflects the relative number of samples associated with that PGCT. Pale blue indicates PGCTs of consisting of control samples only, and the darker shade of purple shows increased number of PreLOS samples in that PGCT. b Temporal change in PGCTs in each individual infant. Red lines represent day of LOS diagnosis. Only samples up to day 50 of life are included. Infant 178 died during the study

Fig. 3

Metabolomic profiles between infants diagnosed with LOS and matched controls across all 5 time points, where TP3 represents samples at diagnosis. a PCA (unconstrained ordination) of LOS infants (red) and matched controls (green). Each sample represented by the small circle and ellipses represent the 95% confidence interval. b Receiver operating characteristic curves of support vector machine predictions for LOS and control samples. AUC represents the strength of the predictive classifications. Selected number of metabolites computed in intervals from 5, 10, 15, 25, 50, and 100 metabolites

Fig. 4

Box plots to show the levels of significant metabolites though each time point between infants diagnosed with late onset sepsis (LOS) and matched controls. Plots listed in order of significance. a Sucrose. b Raffinose. c L-Glutamate. d Didemethylcitalopram. e Acetic acid. f 18-Hydroxycortisol. g 18-Oxocortisol. h L-alpha-Acetyl-N-normethadol

Fig. 5

Spares partial least squared correlations (sPLS) between dominant bacterial genera and identified metabolites. sPLS in regression mode (predict Y from X) to model a causal relationship between bacterial genera and metabolites. Bacterial genera represented in green boxes. Red boxes are metabolites significantly increased in LOS, blue boxed significantly increased in controls, yellow boxes are not significantly altered between LOS and controls. Significant metabolites based on the samples at diagnosis (time point 0)