Center Variation in Intestinal Microbiota Prior to Late-Onset Sepsis in Preterm Infants

Abstract

Objective: Late onset sepsis (LOS) contributes to mortality and morbidity in preterm infants. We tested the hypotheses that microbes causing LOS originate from the gut, and that distortions in the gut microbial community increases subsequent risk of LOS.

Study design: We examined the gut microbial community in prospectively collected stool samples from preterm infants with LOS and an equal number of age-matched controls at two sites (Cincinnati, OH and Birmingham, AL), by sequencing the bacterial 16S rDNA. We confirmed our findings in a subset of infants by whole genome shotgun sequencing, and analyzed the data using R and LEfSe.

Results: Infants with LOS in Cincinnati, as compared to controls, had less abundant Actinobacteria in the first samples after birth (median 18 days before sepsis onset), and less abundant Pseudomonadales in the last samples collected prior to LOS (median 8 days before sepsis onset). Infants with LOS in Birmingham, as compared to controls, had no differences identified in the first sample microbial communities, but Lactobacillales was less abundant in the last samples prior to LOS (median 4 days before sepsis onset). Sequencing identified detectable levels of the sepsis-causative organism in stool samples prior to disease onset for 82% of LOS cases.

Conclusions: Translocation of gut microbes may account for the majority of LOS cases. Distortions in the fecal microbiota occur prior to LOS, but the form of distortion depends on timing and site. The microbial composition of fecal samples does not predict LOS onset in a generalizable fashion.

Fig 1. Barplots of taxa identified as differential abundant in cases and controls.

Y-axis was transformed by taking the base ten logarithm of the number of reads plus one. Samples were rarefied to 2000 reads per sample prior to creating the plots. A) Taxa differentially abundant by LEfSe in the first sample analysis in Birmingham. Median number of reads of Clostridiales in controls was 0, range was 0 to 5. Median number of reads of Clostridiales in cases was 2, the range was 0 to 50. Class Clostridia was also significantly different between cases and controls, it is not shown here because class Clostridia and order Clostridiales had a correlation of 1. B) Taxa differentially abundant by LEfSe in the first sample analysis in Cincinnati. Median number of reads of class Actinobacteria in controls was 1.5, range was 0 to 985. Median number of reads of class Actinobacteria in cases was 0, range was 0 to 74. Median number of reads of Pseudomonadales in controls was 2.5, range was 0 to 661. Median number of reads of Pseudomonadales in cases was 0, range was 0 to 193. Median number of reads of Acinetobacter in controls was 1.5, range was 0 to 660. Median number of reads of Acinetobacter in cases was 0, range was 0 to 75. Median number of reads of Proteus in controls was 0, range was 0 to 267. Median number of reads of Proteus in cases was 0, range was 0 to 1, Phyla Actinobacteria was also significantly different between cases and controls, it is not shown here because it has a correlation 0.9996 with class Actinobacteria and was visually indistinguishable. Family Moraxellaceae was also significantly different between cases and controls; it is not shown here because it had a correlation of 1 with genus Acinetobacter.

Fig 2. First sample classification tree results.

(A) Birmingham results, showing infants colonized with Streptococcaceae and infants with higher levels of Clostridia are at greater risk of sepsis. (B) Cincinnati results, showing that infants colonized with greater relative abundance of Actinobacteria and infants with lower relative abundance of Bacilli are protected from sepsis.

Fig 3. Last sample classification tree results for (A) Birmingham showing infants without Lactobacillales, with the unknown OTU of Bacillales, or born at younger gestational ages were more at risk of sepsis and (B) Cincinnati showing that infants with high levels of Enterobacteriales or high levels of Firmicutes were at increased risk of sepsis.

Fig 4. Barplots of taxa identified as differential abundant in cases and controls at genus level and higher in Cincinnati.

Y-axis was transformed by taking the base ten logarithm of the number of reads plus one. Samples were rarefied to 2000 reads per sample prior to creating the plots. Median number of reads of Pseudomonadales in controls was 0, range was 0 to 1251. Median number of reads of Pseudomonadales in cases was 0, range was 0 to 9. No control samples had reads of Prevotella. Median number of reads of Prevotella in cases was 0, range was 0 to 384. Family Prevotellaceae was also significantly different between cases and controls, it is not shown here because it had a correlation of 1 with genus Prevotella. All other differences identified by LEfSe were at the OTU level. All differences identified by LEfSe in the Birmingham last sample analysis were at the OTU level.