The development of bacterial patterns in preterm infants has not been previously reported. Interactions between the preterm infant and its enteric bacteria are unique. The preterm infant is quite different from the full term infant as the preterm infant possesses an immature gut that must complete development in the extrauterine environment. Additionally, the preterm infant acquires its microbiota within the confines of the neonatal intensive care unit where colonization is significantly influenced by iatrogenic manipulations including: a hospital environment; frequent use of broad spectrum antibiotics, opioids, and histamine-2 receptor antagonists (H2 blockers); and instrumentation with endotracheal tubes, feeding tubes, and suctioning tubes. It is possible that these nosocomial interactions alter the preterm infant’s endogenous microbiota, thereby influencing development of the immature preterm gut and its susceptibility to NEC.
Our data suggest that development of the microbiota in preterm infants differs markedly between patients that do or do not ultimately develop neonatal necrotizing enterocolitis. Unifrac-based PCoA of the 16S rRNA-based 454 sequence data reveals clear temporal separation between the early samples (< 6 weeks of life) in the control preterm infants and those from the healthy full term breast-fed infant along the first principal component axis. These observations suggest a developmental component to both the microbial community as well as host physiology. Interestingly, samples from the controls at ≥ 6 weeks of life, cross the first principal component axis and appear to converge toward the microbiota pattern of the healthy full term infant. If the microbiota of the healthy full term infant represents a necessary complement of microbiota for gut function, clustering of later time point preterm microbiota with this community may represent development towards a native, healthy microbiota.
Samples from preterm control infants of all gestational ages followed the above temporal progression, suggesting that microbiota development occurs over a requisite period of time rather than correlating with a specific gestational age. This microbial community difference may represent the result of clear differences in the environment of the neonatal intensive care unit as outlined above, or an immaturity in the intestinal epithelium. Previous studies have noted that immature intestinal epithelium has decreased intestinal motility, altered surface glycoconjugate glycoslation patterns and altered intestinal mucus density and carbohydrate content, all of which may alter bacterial binding [12, 13].
In contrast, comparison of NEC and control patients demonstrates a distinct trajectory of community development in patients that develop NEC. The ultimate trajectory of microbiota in these infants does not overlap with that of control infants, suggesting that at crucial times a healthy trajectory can be misdirected towards deviant states (for example, NEC). The deviation from control appears to occur two to three weeks prior to NEC. Previous work by others  had noted a similar deviation from healthy infants but only at one week prior to the onset of NEC. Earlier detection, as demonstrated here, would provide an extended timeframe for intervention, but is also indicative of a more prolonged microbial community deviation prior to pathology emergence. Additionally, the earlier detection of this trajectory also led us to sequence metagenomes from earlier time points (three weeks prior to NEC diagnosis) and allowed the identification of functional gene sets that may contribute to the onset of NEC. The initial time point from the set of twins we sampled coincided with the two to three week transition period that we identified in the original control cohort (Figure 2). Although the 16S-based data indicate a shift between controls and those that go on to develop NEC, the first time point comparison between twins, based on ordination of the functional genes detected in the metagenomes does not reflect a major difference. This may be explained by organisms that differ at taxonomic resolutions finer than those reported here, occupying a similar ecological niche between patients at this first time point. In other words, there is a functional redundancy between the community memberships wherein different organisms can perform the same community function. Our present efforts are a coarse level description of functional differences between samples that will provide a candidate set of functions to investigate at a finer scale resolution in subsequent work.
This pattern of deviation prior to onset of NEC is reiterated in our shotgun metagenomic data from twins (Figure 4). NEC may result from the presence of a pathogenic pattern, or alternately, the absence of a beneficial one. Reads from each sample were recruited to available sequenced genomes including candidate pathogens. In all four datasets, the most reads recruited (average of 15.07 ± 2.23% post QC reads) were to the genome from Escherichia coli strain K-12 (Additional file 1: Table S3). Recruited reads to known pathogens (Shigella sp. D9, Citrobacter koseri, Klebsiella pneumoniae subsp. pneumonia, Yersinia pestis, and Cronobacter sakazakii) were substantially lower but consistent across samples (Additional file 1: Table S3) from both patients indicating a diminished role in development of NEC.
To examine differences in the potential function of the microbiota in NEC versus control patients, metagenomic analysis of samples prior to the incidence of NEC was conducted on a pair of twins, one of which went on to develop NEC. Evaluation of this set of twins enabled us to control for environment (neonatal intensive care unit, and diet) and host factors (genetics), allowing us to observe the development of the microbial community over time, with minimal bias introduced by environment or background genetics.
Functions that were statistically significant were dominated by carbohydrate associated genes that mapped to the Enterobacteriaceae (Figure 5). These may represent differences in host metabolism, altering substrate delivery to downstream microbes, or may represent differences in inherent bacterial metabolic pathways that may influence production of metabolites such as short chain fatty acids that affect gut function. Most of the carbohydrate genes identified as significantly distinct between the pre-NEC patient sample and the other three samples were enriched within the pre-NEC dataset.
Only pyruvate carboxyl transferase (EC 220.127.116.11), and two forms of glyoxylate reductase (EC 18.104.22.168; EC 22.214.171.124) were found in lower abundance in the pre-NEC sample. Both of these gene sets are part of the anaplerotic sequences (replenishing reactions) for the tricarboxylic acid cycle to provide precursors for biosynthesis. Pyruvate carboxyl transferase catalyzes the conversion of pyruvate to oxaloacetate while glyoxylate reductase is an enzyme in the oxalate synthesis pathway responsible for converting glyoxylate to glycolate . Alterations in these gene sets may be markers of the health of the microbial community. As part of the anaplerotic reactions, both genes are upregulated when an organism is subsisting on substrates other than glucose.
Microbial communities may metabolize milk through alternative paths, resulting in differential substrate availability for specific microbial groups. The observation of the significantly lower representation of these particular gene sets in the pre-NEC dataset, in combination with an overall enrichment of genes involved in the utilization of carbohydrates, may reflect a change in substrate availability for specific community members presaging the onset of NEC. The pre-NEC community exhibits an expansion and enrichment of these gene sets that were not present earlier, stabilizing its ability to adapt to changes in substrate availability. Furthermore, the distinct profile of functional genes that are enriched in the pre-NEC sample (relative to the three healthy controls) reflects a narrowing of the overall microbial community diversity.
Alterations in these gene sets may also result in bacterial products that have significant effects on the host. For example, a deficiency in pyruvate carboxylase results in conversion of excess pyruvate to lactate. In the infant gut, lactate is used by the intestinal mucosa for gluconeongenesis to supply the small intestinal muscle with glucose . Additionally, a deficiency in glyoxylate reductase may result in increased levels of glyoxylate which has been shown to inhibit the interaction of the Vitamin D receptor (VDR) with vitamin D response elements (VDREs) . Vitamin D has been shown to be involved in the pathogenesis of inflammatory bowel disease and to have immunomodulatory effects .
Genes annotated as cofactor, vitamin, prosthetic group, and pigment subsystems were found to be in higher abundance in the pre-NEC sample. These were primarily associated with tetrapyrroles, specifically, heme and siroheme biosynthesis (precorrin-2 oxidase [EC 126.96.36.199] and sirohydrochlorin ferrochelatase [EC 188.8.131.52]). Precorrin-2 oxidase is a multifunctional enzyme that catalyzes the SAM-dependent methylation of uroporphyrinogen III to form precorrin-2 and trimethylpyrrocorphin 2 and also the subsequent conversion of precorrin-2 into siroheme. Siroheme plays a major role in the sulfur assimilation pathway: converting sulfite to a biologically useful sulfide. The etiology of ulcerative colitis is uncertain but may relate to environmental factors in genetically predisposed individuals. Sulfate-reducing bacteria (SRB) have been implicated through the harmful effects of hydrogen sulfide, a by-product of their respiration, as it is freely permeable to cell membranes and inhibits butyrate oxidation by the host. However, inspection of the taxonomic assignments of the annotated genes (via phylogenomic inference) across all four samples revealed a lower relative abundance of known SRBs when compared to the other three samples, indicating that a sulfide-mediated phenomenon may not be a key requirement for the onset of NEC.
Within the virulence, disease and defense subsystems, genes encoding for resistance to antibiotics and toxic compounds were detected within both the Group B and Group D gene sets. Those detected in the Group B set (over-represented in the pre-NEC sample) are part of the mdtABCD multidrug resistance gene cluster (encoding for the multidrug transporter MdtC), which confer resistance to deoxycholate and novobiocin. In the Group D set, which contains those genes that are under-represented in the pre-NEC sample, the resistance genes detected are related to the BlaR1 protein family (specifically peptidase M48, Ste24p precursor). This group of genes encodes a family of membrane-spanning proteins with a penicillin sensor and a signal transducer domain that provides the starting point for this resistance mechanism. While utilization of neither antibiotic is noted in the clinical records for the patient, deoxycholate is a bile salt and may be indicative of a community response to changes in the intestinal contents.
It has been suggested that premature infants would benefit from optimized intestinal bacterial colonization; however ‘optimal’ colonization has not been defined. Our data suggest that the colonization pattern of the healthy full term breast-fed infant may represent this optimal colonization for the preterm infant as well. While several studies have evaluated the progression of colonization patterns in healthy full term infants, and have demonstrated that there is high variability between colonization patterns of individual infants at early ages, some patterns have emerged . Previous sequencing based studies have demonstrated that immediately after vaginal delivery, infants have colonization patterns reflective of the maternal vaginal compartment including Lactobacillus, Prevotella, or Sneathia species . Culture-based studies have then shown that these infants have earlier colonization with Bifidobacterium and Lactobacillus. Breast-feeding is associated with colonization by Bifidobacterium, Lactobacillus and Streptococcus[21–25]. Several clinical studies have shown a beneficial effect of probiotics on the incidence of NEC [26–28]. In these studies, probiotic administration was begun early with primarily Bifidobacteria and Lactobacillus species that may have altered the overall community toward the healthy breast-fed infant community profile.
Our data suggest that microbial community developmental patterns are important for health. It is possible that early patterns induce key developmental pathways critical for protection of the preterm infant. Studies have shown that bacteria can influence maturation in many ways - prevention of apoptosis, improvement of intestinal barrier function, establishment of intestinal mucus, maturation of glycoconjugate patterns, improvement in immune response, and maturation of intestinal vasculature patterns. All of these may be important for health and protection against NEC. Further testing beyond the scope of this study is necessary to confirm the mechanistic implications of the microbiota differences found for NEC pathogenesis. The host response to these microbiota differences, also warrants further study. However, since bacterial patterns differ between control and NEC patients, our data suggests that bacterial therapy may be an option for NEC prevention. Our studies suggest that a shift toward the microbial patterns seen in control infants may be protective. Additionally, our data suggest that identifying a shift in microbial community three weeks prior to NEC could be used to predict patients at risk for NEC. As such, our data emphasizes the importance of the temporal nature of the microbial community development in preterm infants. Identifying the critical time point, key microbial components, and functions will be necessary for effective bacterial therapeutics to prevent NEC. In the neonatal population, microbial patterns are not yet established; thus, in this patient group one has the unique opportunity to initiate and influence colonization. Our data demonstrate a difference in microbial community structure by both taxonomy and function several weeks prior to NEC. This suggests a window of opportunity in which interventions to alter community structure could influence clinical outcome.