704 - Associations between taxonomic composition and absolute bacterial abundance in the preterm gut microbiome

Publication Number: 1681.704

Naomi G. Wilson, Childrens Hospital of Philadelphia, St Louis, MO, United States; Miren B. Dhudasia, Childrens Hospital of Philadelphia, Philadelphia, PA, United States; Kyle Bittinger, Childrens Hospital of Philadelphia, Philadelphia, PA, United States; Sagori Mukhopadhyay, Childrens Hospital of Philadelphia, Philadelphia, PA, United States

Background: The preterm gut microbiome demonstrates dynamic age-related shifts in composition. While these relative changes have been described, few studies have examined how absolute bacterial abundances change with age or how it correlated with relative abundance.

Objective: To quantify age-related changes in absolute bacterial abundance in very low birth weight (VLBW, < 1500 gram birthweight) infants and correlate it with relative taxonomic abundance.

Design/Methods: We conducted a single-center prospective cohort study of 101 VLBW infants born ≤34 weeks and enrolled after parenteral consent. Fecal samples were collected weekly from birth until hospital discharge and analyzed by shotgun metagenomic sequencing. Each infant contributed a median of 8 samples (IQR 6-11), totaling 904 samples. Absolute bacterial abundance was quantified using 16S rRNA gene qPCR and expressed as genome copies per µL. Trajectory of absolute bacterial abundance with age was examined using segmented breakpoint analysis and correlations between relative and absolute abundance was assessed with t-tests and a Gaussian Mixture Model.

Results: Median (IQR) gestational age and birthweight of the cohort was 28.4 (26.9-29.9) weeks and 1105 (845-1295) grams; 49% were female and 75% were born by C-section. Most (92%) received human milk for >75% of hospitalized days. The total fecal bacterial load rose rapidly during the first 8 days, then stabilized. Segmented breakpoint analysis identified one breakpoint at 8 days age (p < 0.001, 95% CI: 7.5-8.5; Figure 1). As is well-described in preterm infants, most samples were dominated by few taxa: Klebsiella (37.8%), Escherichia (17.3%) and Staphylococcus (9.2%). After excluding early samples ( < 8 days age) where low bacterial load maybe related to age, samples with lower than the average bacterial load were significantly more often Staphylococcus-dominant (Figure 2). Contamination is unlikely as the total load of these samples were significantly higher than negative controls. A smaller subset of low bacterial load samples (≈55±4 genome copies per µl; Figure 3, red curve) were Klebsiella-dominant, though on average Klebsiella-dominant samples had higher bacterial loads (green curve).

Conclusion(s): The association of Staphylococcus and Klebsiella with low bacterial load samples suggests possible recolonization patterns after interventions in the NICU that reduce bacterial load, such as antibiotic exposure or fasting. These patterns may influence future health outcomes, such as the risk of infections. Understanding these relationships can help promote microbiome recovery patterns associated with health benefits.

Figure 1: Stool bacterial load in preterm infants rapidly increases in first week of life and afterwards remains the same
Bacterial load in log-scale of the stool samples is shown on the y-axis, with days of life of the infant at stool sampling on the x-axis. Each point represents one stool sample from one individual. Segmented breakpoint analysis performed, and one breakpoint was found, marked with a dashed blue line and blue p-value. When fitting a linear model to the log-transformed bacterial load values, R package segmented::segmented selected a breakpoint at 8 days of life (dashed blue line) at p<0.0001 with a 95% confidence interval 7.5-8.5.

Figure 2: Staphylococcus dominated samples have low bacterial load
Samples obtained ≥8 days were grouped by their highest relative abundance genus of at least 25% and each group's bacterial load was compared to the global mean with a t-test.

Figure 3: A subset of Klebsiella-dominated samples have low bacterial load.
A 2-component Gaussian Mixture Model was fit to the data: the red curve traces the density fit of the low bacterial load samples and the green curve traces the density fit of the higher bacterial load samples. The relative likelihood of the data fitting better to a single normal distribution compared to a 2-component GMM is very low (7.9E-64). X-axis is the log-transformed estimated genome copies per ul in Klebsiella-dominated fecal samples