Intestinal colonization has long been thought to contribute to NEC in preterm infants, despite a failure to consistently identify a single pathogen or pathogenic microbial community associated with its occurrence. The results of this study indicate the need to focus metagenomic research for preterm infants on the first few weeks of life. Our study revealed several factors that were significantly associated with subsequent risk of NEC, specifically, the absence of Propionibacterium in the first week and two distinct forms of dysbiosis that occurred over the first two weeks of life: During days 4 to 9, the microbial community prior to onset of early NEC cases consisted predominantly (≥98%) of Firmicutes, specifically class Bacilli, of which the dominant genera were Staphylococcus and Enterococcus. This unique Gram-positive microbial signature was not shared by any of the control infants. The second clustered microbial phenotype occurred during 10 to 16 days of life preceding later onset NEC cases, and consisted of a Gram-negative Proteobacteria signature, specifically, family Enterobacteriaceae, of which the dominant genera were Enterobacter and Escherichia. The disparate timing and composition of these high Firmicutes and high Proteobacteria microbial signatures are intriguing, especially given their association with early versus late onset NEC or death. We also found maternal antepartum antibiotic use to be significantly higher in NEC-II compared to NEC-I cases. While there appeared to be a trend towards association of NEC-I with more C-section delivery and NEC-II with more vaginal delivery, this was not significant. Our data also suggested a trend towards lower alpha diversity of samples from later NEC cases compared to controls, but this was also not statistically significant. However, the limited sample size of this study does not preclude the possible differences that might be identified from analysis of larger studies of the early microbiome in future.
While we are not aware of any previous study that has demonstrated two distinct forms of intestinal dysbiosis prior to onset of NEC, previous studies support the plausibility of our findings. Consistent with our finding that distinct forms of dysbiosis were composed of Firmicutes and Proteobacteria, Koenig et al.  reported that these phyla are strongly negatively correlated with each other in normal infant colonization. A previous study of 18 preterm infants by Mai et al.  reported that the relative abundance of Firmicutes was higher in samples taken one week prior to NEC onset compared to controls, followed by higher abundance of Proteobacteria within 72 hours prior to NEC onset. Although the authors did not describe distinct forms of dysbiosis for distinct subgroups as reported here, their findings nevertheless suggested a key role for both major phyla at distinct timings. Finally, Taur et al. reported two forms of dysbiosis following antibiotic administration in allogeneic hematopoietic transplant patients; consistent with our study, one form of dysbiosis was characterized by dominance of Firmicutes (Enterococcus and Streptococcus) and the other form was characterized by dominance of various Proteobacteria .
Our discovery of two forms of dysbiosis by metagenomic analysis was supported by metabolomics, which identified differences among urine samples that were collected in the first week of life prior to case onset. These two ‘-omic’ methods provide complementary information. Urinary metabolomics is a sensitive method of identifying groups that differ in their intestinal bacterial colonization [25, 44]. Production and utilization of specific metabolites differ among colonizing bacteria, which in turn affects their bioavailability to the host . While metagenomic analysis of microbial DNA provides a comprehensive snapshot of bacterial composition, metabolomic comparison of microbial colonization phenotypes provides a snapshot of their differential metabolic activity [27–29]. In our study, three urinary metabolites (alanine, pyridoxine and histidine) differed significantly between one of the NEC sub-types and controls. Urinary alanine was higher in NEC-I compared to controls, positively correlated with the relative abundance of Firmicutes, and negatively correlated with the relative abundance of Proteobacteria and Propionibacterium. Alanine is a non-essential amino acid, which is ubiquitously incorporated into bacterial cell wall biosynthesis, a potential target of immune sensing [58–60]. As peptidoglycan constitutes most of the dry weight of Gram-positive organisms but only a small share of the dry weight of Gram-negative organisms, alanine seems a particularly promising candidate to differentiate patients whose microbiome is strongly dominated by Gram-positive organisms. Indeed, alanine was previously reported to be significantly elevated in the feces of irritable bowel syndrome cases compared to controls, and positively associated with higher intestinal colonization with Gram-positive organisms . Pyridoxine, also elevated in the urine of NEC-I infants versus controls, is produced by bacteria in general  and may reflect bacterial abundance or growth. But pyridoxine, a correlate of alanine in our data, did not appear to be independently associated with microbial community composition or NEC in the presence of alanine. A proteinogenic amino acid, histidine, differed between NEC-II and controls, but was not associated in our dataset with microbial community composition per se. However, an in vitro study of the metabolites of microbial growth previously reported that histidine was lower in medium with increased growth of Escherichia, a genus that contributed to the Proteobacteria dysbiosis associated with NEC-II in our study . While none of the metabolites were alone predictive of overall NEC, the ratio of alanine to histidine was significantly associated with NEC overall as well as with the relative abundance of Propionibacterium.
The preterm infants in our study generally lacked microbiota that are known to influence healthy immune development and oral tolerance, including Bifidobacterium, Bacteroides fragilis and other commensal gut microflora [23, 24, 62]. While the lack of these beneficial organisms is characteristic of preterm infants, the primers used in this study have been documented to be less than optimal for quantitative representation of Bifidobacterium. However, our own unpublished data suggest that the presence or absence of Bifidobacterium was unlikely to have been highly biased. In this study, we detected Bifidobacterium in 15% of the samples. From the same ongoing cohort of preterm infants, we recently analyzed an additional 182 samples that were amplified using 515 F/806R primers and sequenced on the MiSeq platform  and again found a 15% Bifidobacterium detection rate. Of the additional samples sequenced both by 515 F/806R (MiSeq) and V3-V5 (454), the same Bifidobacterium-positive samples were identified with each method. For one Bifidobacterium-positive sample, we generated additional data from deep whole genome sequencing (WGS). The estimated relative abundances of Bifidobacterium by method were 22% by WGS, 14% by 515 F/806R (MiSeq) and 2% by V3-V5 (454). Taken together, these data suggest that our use of V3-V5 primers did not significantly misrepresent the presence or absence of Bifidobacterium colonization in our dataset, but very likely underrepresented the relative abundance of Bifidobacterium when present.
Propionibacterium, a genus of the phylum Actinobacteria, was the only organism that differed significantly between all NEC cases and controls in this study. The organism was identified in the first postnatal week in about half of the controls but none of later NEC cases, suggesting a potential commensal role. This genus includes many species and strains that are used as probiotics by the dairy industry . Other Propionibacterium commonly colonize the skin  and have been reported in breast milk . These organisms are so named due to their production of propionic acid as well as other short chain fatty acids that have a beneficial role in intestinal health. The role of Propionibacterium in the intestinal colonization of infants is not known. Nevertheless, our observation suggests a benefit from initial colonization with this organism, and the possibility that other commensals may also benefit preterm infants.
The high Firmicutes dysbiosis that we observed may imply excessive exposure to the peptidoglycan that covers the surface of Gram-positive organisms. TLR2 recognizes peptidoglycan, and exhibits excessive signaling in the immature enterocyte . The lack of exposure to LPS-bearing Gram-negative organisms in the first week of life may impair the development of tolerance in preterm infants, resulting in an even higher inflammatory response when presented. The Proteobacteria dysbiosis that we observed was characterized by high abundance of Enterobacteriaceae in the second week of life, consistent with in vivo and ex vivo studies indicating that NEC is a hyperinflammatory state resulting from excessive TLR4 signaling in response to LPS [9–11, 13]. However, in our study, high Gram-negative predominance occurred in both NEC and control subjects. High relative abundance of Proteobacteria may induce heightened vulnerability, but the development of NEC may require additional insults or vulnerabilities, including later exposures to pathogens or oxidative stress . Alternatively, control infants with similarly high colonization with Proteobacteria may be more immunologically tolerant. Our finding that no NEC cases but half of the controls had low but detectable levels of Propionibacterium, and that the presence of this organism appeared to mitigate the risk of NEC associated with high levels of Proteobacteria in the second week of life, is consistent with the concept that the early presence of commensal bacteria helps induce immune homeostasis .
While early colonization might contribute in various ways to the causal pathway leading to NEC, our findings are consistent with the suggestion that early dysbiosis has a time-sensitive role in dysregulation of the developing immune response [19, 21]. Supporting this putative role, a time-series analysis of the intestinal transcripts of gnotobiotic and conventionalized mice demonstrated that immune genes respond to microbial colonization in a temporal sequence that coordinates the development of the immune system to achieve homeostasis . For example, T-cell maturation and tolerance-associated functions, such as IL-10 and Foxp3, are significantly increased in conventionalized mice by day eight after introduction of microbiota, a timing relevant to our findings. However, our study does not address causality. Experimental studies are needed of microbial host interactions and immune development of the immature mucosa to understand the potential role of time-specific microbial dysbiosis as related to NEC in preterm infants.
The limitations of this study should also be considered. The sample size of this study was modest, and our findings require validation in larger, multi-site cohorts. Further, we included three non-NEC deaths as a secondary comparison group. The cause of death for the three non-NEC deaths was attributed to respiratory distress syndrome for two infants and suspected infection for one infant. While these deaths may be considered irrelevant to NEC, our data suggest similarity in the microbiome of NEC and the non-NEC (non-congenital) neonatal deaths that occurred in the same postnatal period [7, 8, 15]. The reason for the observed microbial similarity between high Firmicutes-associated NEC and non-NEC deaths in this study is not known. None of the non-NEC deaths were ever suspected of NEC. The infant with suspected infection had clinical signs of infection and received antibiotics and supportive treatment, but blood cultures were negative. Since the pathobiology of NEC remains an enigma, and no pathogens were identified that may have contributed to the non-NEC deaths, we cannot explain the similarity of the microbiome in non-NEC deaths and early NEC cases. Nevertheless, these initial findings suggest that immune dysregulation secondary to dysbiosis may be common to NEC and some non-NEC deaths.