The Metabolome and the Gut Microbiota for the Prediction of Necrotizing Enterocolitis and Spontaneous Intestinal Perforation: A Systematic Review

Necrotizing enterocolitis (NEC) is the most devastating gastrointestinal emergency in preterm neonates. Research on early predictive biomarkers is fundamental. This is a systematic review of studies applying untargeted metabolomics and gut microbiota analysis to evaluate the differences between neonates affected by NEC (Bell’s stage II or III), and/or by spontaneous intestinal perforation (SIP) versus healthy controls. Five studies applying metabolomics (43 cases, 95 preterm controls) and 20 applying gut microbiota analysis (254 cases, 651 preterm controls, 22 term controls) were selected. Metabolomic studies utilized NMR spectroscopy or mass spectrometry. An early urinary alanine/histidine ratio >4 showed good sensitivity and predictive value for NEC in one study. Samples collected in proximity to NEC diagnosis demonstrated variable pathways potentially related to NEC. In studies applying untargeted gut microbiota analysis, the sequencing of the V3–V4 or V3 to V5 regions of the 16S rRNA was the most used technique. At phylum level, NEC specimens were characterized by increased relative abundance of Proteobacteria compared to controls. At genus level, pre-NEC samples were characterized by a lack or decreased abundance of Bifidobacterium. Finally, at the species level Bacteroides dorei, Clostridium perfringens and perfringens-like strains dominated early NEC specimens, whereas Clostridium butyricum, neonatale and Propionibacterium acnei those at disease diagnosis. Six studies found a lower Shannon diversity index in cases than controls. A clear separation of cases from controls emerged based on UniFrac metrics in five out of seven studies. Importantly, no studies compared NEC versus SIP. Untargeted metabolomics and gut microbiota analysis are interrelated strategies to investigate NEC pathophysiology and identify potential biomarkers. Expression of quantitative measurements, data sharing via biorepositories and validation studies are fundamental to guarantee consistent comparison of results.


Introduction
Necrotizing enterocolitis (NEC) is the most devastating gastrointestinal emergency in preterm neonates, which still carries a high burden of neonatal morbidity and mortality worldwide. NEC is estimated to affect about 6% (3-9%) of very low birth weight infants (birth weight less or as 1500 g) [1][2][3], and its mortality rate can reach 50.9% in extremely low birth weight infants (<1000 g) with surgical disease [3]. NEC is related to several short-

Inclusion Criteria
Titles were screened for relevance and duplications by three reviewers independently, with disagreements resolved by discussion. Cross-sectional studies and studies with "prospective-specimen-collection, retrospective-evaluation" (PRoBE) [20] (single-group or multiple-groups) were included if they evaluated neonatal metabolome and/or gut microbiota in preterm infants with NEC and/or SIP, compared to healthy controls. Only studies applying untargeted techniques and culture-independent molecular techniques were included. The definition of NEC according to Bell's stage II or III was considered for qualitative analysis. Spontaneous intestinal perforation was defined as isolated intestinal perforation with the presence of free intraperitoneal air in the absence of pneumatosis intestinalis [21]. Whenever possible, results were assessed and interpreted according to the timing of samples' collection: early-samples, collected in the first weeks of life and >72 h from NEC onset (NEC prediction); late-samples, collected within 72 h from NEC onset (NEC diagnosis). Case series, brief report, unpublished studies (e.g., conference abstracts), and animal studies, were excluded. Abstracts for which the full text was in a language other than English were excluded. In the case of unavailable full text/data, authors were contacted directly by the reviewers.

Outcomes
The primary objective was to determine the differences in metabolic and gut microbiota patterns between infants with NEC Bell's stage II or III and healthy controls. As secondary outcomes we evaluated the differences in metabolic and gut microbiota patterns between infants with NEC and those with SIP, and between lethal NEC (NEC-related death) and NEC not associated with mortality.

Data Extraction and Synthesis
The results of studies applying metabolomic analysis were expressed as absolute or relative presence (% of subjects), fold change/fold change ratio (FC/FCR), area under the curve (AUC), relative intensity (RI) or peak intensity (PI), or arbitrary unit (AU), whenever possible. The results of studies applying untargeted gut microbiota analysis were expressed as absolute or relative presence (% of subjects), relative abundance (RA), reads, or odds ratio (OR), when these data were available. Indices of alpha diversity within samples through time as expressed by species richness (Chao-index or Chao-i), or by richness and evenness estimators (Simpson and Shannon diversity indices) [22,23] were collected, whenever mentioned. Beta-diversity indices, as measures of between samples diversity and phylogenetic distance, expressed by weighted and unweighted UniFrac metrics [24] were considered as well, when available.

Results
A total of 1191 studies were screened for full text extraction after the removal of 811 duplicates. Of the full texts evaluated, 1158 were excluded due to valid reasons (title and abstract screening, missing data, unavailable full-text, or for not including the intervention or outcome of interest, Figure 1). Thirty-three studies were selected. Six studies applied untargeted metabolomic analysis, 23 used untargeted microbiota analytic techniques, and four applied both (Tables 1 and 2) . Only five studies applying metabolomics and 20 studies evaluating gut microbiota were considered for qualitative analysis, as they applied the strict NEC definition of Bell's stage II or III. Figure 1 shows the PRISMA flow diagram of the study. The PRISMA checklist is available as Supplementary Material. Table 3 [58][59][60][61][62][63][64][65][66][67][68] reports the characteristics of the excluded studies after eligibility evaluation.   Table 1. Characteristics of studies applying untargeted metabolomic analysis for the predisposition/diagnosis of NEC/SIP (NEC/SIP prediction: early samples, first 14 days or >72 h from NEC onset; NEC diagnosis: late samples, <72 h from NEC onset). In grey shadow the studies which used NEC Bell's stage II or III as definition and which were therefore included in the review. Numerical data are expressed as mean (SD) or median (IQR) or as number/percentage, if not otherwise specified. Abbreviations: ANS = antenatal steroids; ATB = antibiotics; AU = arbitrary unit; AUC = area under the curve; BRM = broad range metabolomics; CAM = chorioamnionitis; CCC = complex congenital cardiopathy; CS = cesarean section; EBM = exclusive breast milk; FC = fold change; FCR = fold change ratio; F = French; FI = feeding intolerance; g = grams; GA = gestational age; GDT = good digestive tolerance; GW = gestational weeks; h = hours; MP = multiple pregnancy; NEC = necrotizing enterocolitis; NI = no information; NMR = nuclear magnetic resonance; N,N-DMG = N,N-Dimethylglycine; PI = peak intensity; RI = relative intensity; SIP = spontaneous intestinal perforation; TP = timepoint; VIP = variable importance plot; VLBW = very low birth weight; w = weeks.
Two studies included early-collected samples (>72 h before NEC onset) [26,29], which were considered as potentially predictive of NEC. Morrow et al. [26] found no single metabolite associated with NEC cases in samples between 4 and 9 days of life, but reported a urinary alanine/histidine ratio >4 to have a sensitivity of 82% with a predictive value of 78% for NEC. The other studies did not find any difference in terms of metabolomic profiles between cases and controls.
Four studies included samples collected in proximity to NEC diagnosis (late-collected samples, within 72 h from NEC onset) [27][28][29]31]. Stewart et al. [29] showed five metabolites in stools of NEC cases having the highest variable importance plot score. These belonged to linoleate metabolism, C21-steroid hormone biosynthesis, leukotriene metabolism, and formation of prostaglandin from arachidonate. Rusconi et al. [28] found significant changes in the sphingolipid pathway between cases and controls, in particular with increased sphingomyelins and decreased ceramides in faecal samples of NEC cases before disease onset. This trend was confirmed by the targeted analysis and was characteristic of NEC Bell's stage II and III.
In urine, Picaud et al. [27] reported significantly decreased relative intensity of N,Ndimethylglycine (N,N-DMG), betaine, creatinine and urea in three cases of late-onset NEC compared to controls. Similarly, Thoimadou et al. [31] identified 14 endogenous metabolites (mainly amino acids and organic acids) to be significantly lower in the urinary profiles of NEC neonates, with succinate, citrate, 4-hydroxybenzoate, proline, tyrosine, and fumarate, having the highest AUC in the ROC curves (0.89, 0.85, 086, 0.83, 0.80, and 0.82, respectively). Tyrosine, fumarate, and proline, were also discriminating metabolites in the targeted LC-MS/MS analysis. The authors identified a multi-biomarker based on tyrosine, arginine, and riboflavin, as having the highest diagnostic ability in discriminating NEC neonates from controls (AUC 0.963).
None of the studies evaluated the difference in metabolic profiles between cases affected by NEC and those affected by SIP.
At the phylum level, five main taxa appeared to be significantly different in NEC Bell's stage ≥II cases compared to controls. Proteobacteria showed a higher RA in NEC cases compared to controls (61-90.7% vs. 19-56.4%), from 2 weeks before NEC [53] up to near or at disease onset [43,45,54]. In controls, Lindberg et al. [43] showed a significant shift towards a reduction in Proteobacteria abundance from the early time in the hospital course to later time points (corresponding to NEC diagnosis). Four studies, instead, did not reveal any significant difference in Proteobacteria abundance between the two groups [35,39,42,46].
Near or at NEC diagnosis, NEC Bell's stage ≥II cases were characterized by a significant lower relative abundance of the phyla Firmicutes [43,54] [26] demonstrated an opposite relationship with lower RA of Proteobacteria and higher RA of Firmicutes in cases early in life (4-9 DOL), and Mai et al. [45] also found predominance of Firmicutes in NEC at the earlier time point.
At the class level, included studies showed a wider variety of results. In two studies [50,57], Clostridia were more represented in NEC cases, whereas in the other two [46,56] this class, alone or combined with Negativicutes, was expressed in lower proportion compared to healthy controls. Nevertheless, in a multivariate analysis, Sim et al. [50] found that the reads for Clostridium operational taxonomic unit (OTU) had an OR of 6.2 (85%CI, 1.8-21.7, p = 0.004) in discriminating NEC Bell's stage II and III from controls.
Contrasting results emerged in relation to the abundance of Gammaproteobacteria, as well, with two studies reporting an increased proportion of this class in cases at the time of NEC onset [54,56], whereas another highlighted the opposite at two weeks of life [57].
No conclusions could be deduced at order and family levels. At the genus level, NEC cases appeared to have higher levels of Bacillus [44] at an early post-partum stage, and of Pasteurella at 2 weeks of life [57]. Controls were characterized by higher abundances of the genera Staphylococcus [40], Propionibacterium [26], and Enterococcus [43,44], in this period of life, instead. In pre-NEC samples, Stewart et al. [29] demonstrated that a preterm gut community type characterized by high RA of Bifidobacterium could not be found in any sample from NEC patients. Similarly, Torrazza et al. [53] showed significantly lower numbers of Bifidobacteria in cases than in controls during the weeks before diagnosis. In near NEC onset, affected infants showed a slight increase in Clostridium sensu strictu [49], Enterobacter [51,57], Staphylococcus [51], E. Coli [55,57], and Shigella [57], and less Veillonella [55]. Streptococcus [46,51], Enterococcus [41,47,51], and Veillonella [43,46], were greater in healthy subjects. In general, studies which described the progression of gut microbiota through different timepoints demonstrated specific trends rather than single putative pathogens before NEC onset. Interestingly, Zhou et al. [57] revealed different changes according to the timing of NEC development (early vs. late).
No studies compared the stool microbiota of NEC infants to those of preterm infants affected by SIP and controls.
As regards alpha diversity within samples through time as expressed by species richness (Chao-1 index), four studies did not find any difference between cases and controls [26,39,45,53], whereas one study reported lower Chao-i in NEC 1-5 days prior to symptoms [46], and in particular in those with lethal NEC. Concerning alpha diversity as expressed by richness and evenness estimators, the Simpson index did not differ between cases and controls in three studies in our review [26,40,43]. The Shannon diversity index, instead, appeared to be significantly lower in NEC Bell's stage ≥II than controls, both early in life and at disease onset, as reported by six of the studies [35,44,46,54,56,57]. Two studies found no significant difference in the Shannon diversity index between the two groups [47,55]. One study found a reduced temporal development of this index in NEC samples, compared to an increase in controls [29].
Studies which performed principal coordinate analyses (PCoA) based on the weighted or unweighted UniFrac metrics showed a clear separation of NEC cases from controls early after birth [44], 1 or 2 weeks prior to NEC [45,53], and close to NEC diagnosis [44][45][46]54], indicating the presence of a certain phylogenetic distance between microbial communities in the two groups. Feng et al. [39] and Normann et al. [47], instead, did not find any clustering by case status.

Combined Untargeted Metabolomics and Gut Microbiota Analysis
Four studies [25,26,29,32] explored both metabolomic and gut microbiota profiles to highlight potential interrelationships between the patient's metabolic pathways and the changes induced by gut microbial activity. However, only two studies [26,29] included patients with stage II or III according to modified Bell's criteria. In the study by Morrow et al. [26], alanine was positively associated with early-onset NEC cases who were preceded by Firmicutes dysbiosis. Alanine was also directly correlated with the relative abundance of this phylum. Histidine, instead, was inversely associated with NEC cases preceded by Proteobacteria dysbiosis and had a strong inverse association with the interval between collection of the urine sample and case onset. The urinary alanine:histidine ratio was inversely associated with the relative abundance of Propionibacterium, which lacked in all NEC samples. Stewart et al. [29] grouped gut microbiota profiles into six clusters named preterm gut community types (PGCTs) as reflected by the dominant core OTUs. The sixth PGCT, dominated by Bifidobacteria and comprising exclusively healthy samples, was the one with the lowest abundance of metabolites associated with NEC.

Discussion
NEC remains one of the most common and yet inexplicable causes of death in preterm infants [69], therefore the search for its non-invasive prognostic and diagnostic biomarkers is one of the most intriguing topics in neonatology and has been summarized in several systematic reviews so far [12,[70][71][72][73][74]. The multifactorial pathogenesis of this complex condition makes it challenging to unravel a single potential biomarker of the disease. The diagnosis of NEC still remains predominantly clinical, with few laboratory and radiological findings as surrogates to identify the severity. Rapid, bedside, point-of-care tests, to be performed prior to or at clinical manifestations, may help in guiding the management, for instance in the matter of feeding strategies, appropriate administration of antibiotics, or the need for urgent surgical intervention [14]. In fact, once the clinical and radiological picture are clearly evident, it may be too late to prevent disease progression and associated catastrophic outcomes [75].
With their hypothesis-free hypothesis-generating approach, the omics technologies may untangle a better understanding of the molecular processes responsible for NEC. The rise in the interest of omic approaches is demonstrated by the exponential increase in "omic" papers in the last 20 years [71,76]. Untargeted techniques applied to NEC have interrogated neonatal serum, urine, and stools, as easily collected biological samples and feasible to explore [10,13,14,71,75,77]. Metabolomics has several advantages over the other omics approaches by detecting the functional end points of cellular reactions and the direct result of a biochemical response to a stimulus [10,76]. As last downstream products of gene transcription and enzymatic pathways, metabolites provide a closer picture of the organism's phenotype and its interaction with the environment. They can be virtually detected in any type of biological specimen and assayed by 1H-Nuclear Magnetic Resonance (1H-NMR) spectroscopy or mass spectrometry (MS). The former exploits the local magnetic field around a nucleus to elucidate the molecular structure of metabolites, whereas the latter separates metabolites using one of a variety of methods, including ultraperformance liquid chromatography (UPLC-MS) and gas chromatography (GC-MS) [72,76,77]. NMR spectroscopy has a relatively lower sensitivity and can detect a smaller number of metabolites compared to MS, but it needs minimal sample preparation, has a high reproducibility, and can be applied on tissue samples directly (for example intestinal tissue). MS, instead, owns a high sensitivity, proven an adequate sample preparation, detecting up to 5000 metabolites, but it has moderate reproducibility and requires a more demanding preparation of samples [78]. Additionally, it needs known standards to exactly identify metabolites of interest. These techniques are analytical distinct but complementary, allowing a wide and sensitive detection of functional low molecular weight metabolites [71]. While blood samples (serum) may better reflect the pathophysiological-metabolic changes of global events, such as NEC, they have the drawback of being invasively collected, as well as potentially associated with iatrogenic anemia [31]. Urine is a suitable non-invasive sample option, however, the accumulation of metabolic products depends on creatinine clearance, which varies according to gestational and postnatal age. Ideally, stool would provide a non-invasive mean which may better reflect the host-microbiota interaction in the gut lumen, the site of interest for NEC [75,77]. Nevertheless, faeces are usually not regularly produced by extremely preterm infants, especially in the first days of life, their composition is very complex and heterogeneous, rich in macromolecules and food-derived metabolites, making their analysis difficult with instrumental methods [75,79,80]. Interestingly, one study explored dried blood spots as a medium in NEC research [65]. However, this was not included in our review due to the application of a targeted approach to amino acids and acylcarnitines. Given the benefits and drawbacks of each biological fluid, it would be reasonable to explore urine, stools, and a small amount of plasma, or dried blood spots, in order to gather the majority of information and to avoid iatrogenic anemia.
Studies applying metabolomics in infants with NEC have been summarized by several reviews in the last 10 years [10,71,72,76,81]. As already highlighted by previous authors, there is a wide variability in populations' inclusion criteria, timing of samples' collection and type of analyzed biological fluids. In our review, only five metabolomic studies applied a rigorous definition of NEC. Although there does not appear to be a unifying metabolomic signature of NEC, some studies show interesting results with involvement of pathways related to inflammatory response [26,29], intestinal permeability [28], and energy depletion, potentially due to an inflammatory state [31].
Stewart et al. [29] were among the first to integrate sequence and metabolomic stool analysis in preterm neonates for NEC. They found five metabolites with a temporal increase prior to NEC diagnosis and to be discriminatory in the NEC samples at the time of diagnosis. These belonged predominantly to the linoleate metabolism, which may be involved in an inflammatory-mediated damage [81]. Another metabolite which may be related to inflammatory changes, lactate, was associated to late-onset NEC cases in Picaud et al. [27]. Intriguingly, the enzyme lactate dehydrogenase was overabundant in stools of infants prior to NEC development in the shotgun metagenomic analysis by Tarracchini et al. [82], supporting the accumulation of lactate in these subjects possibly due to an imbalance between lactate-producing and lactate-utilizing bacteria. Rusconi et al. [28] were the only group to validate their initial untargeted observations in a targeted analysis of samples on an expanded, not fully independent, cohort. They showed significantly increased sphingomyelins and decreased ceramides in pre-NEC stools of Bell's stage II and III cases versus Bell's stage I and controls. The authors attributed these changes to a decrease degradation of sphingolipids, as they were not related to dietary differences. These are bactericidal components of cell membranes against gram bacteria, and they have been hypothesized to be relevant to NEC development as they can alter microbiome composition [83].
Finally, Thoimadou et al. [31] highlighted differences in several metabolic pathways between NEC cases and controls, with the highest impact expressed by amino acids' metabolism. The reduction in some amino acids in cases, namely alanine, asparagine, and proline, was hypothesized to promote the Krebs cycle in order to maintain enterocytes' supply energy and integrity during an inflammatory state. The low urinary alanine level in NEC is apparently in contrast with the results by Morrow et al. [26], but the timing of sample collection differed between the two studies, and findings by Morrow et al. were strictly related to the microbiome dysbiosis dominated by Firmicutes. An early metabolic deviation involving predominantly amino acids has also been reported by Sinclair et al. [65], who found reduced levels of tyrosine and increased levels of its precursor, phenylalanine, in NEC, similarly to Thoimadou and colleagues [31]. Both authors explain this alteration with a potential impairment in the conversion of phenylalanine into tyrosine occurring in critically ill neonates during catabolic and oxidative stress states [31,65]. Interestingly, similarly to Rusconi et al. [28], Thoimadou and colleagues [31] did not demonstrate any possible influence of NEC stage I on the LC-MS data, consistent with the thinking that stage I NEC may be more similar to feeding intolerance than Stage II and III NEC.
The role of microbial dysbiosis prior to NEC is supported by the fact that NEC cannot be produced in germ free animals, as well as by the positive association between antibiotic use and the disease [84,85]. Animal studies suggest that microbiota composition in the neonatal period may affect gastrointestinal (GI) tract development, mucosal integrity, and even nutritional status [86,87]. Additionally, although NEC does not occur in utero, meconium is not sterile and its microbial composition, most likely reflecting the in-utero environment, varies depending on gestational age, with the potential involvement in the subsequent development of sepsis and NEC [40,81,88]. After birth, neonatal microbiota is influenced by mode of delivery, early nutrition, type of nutrition, as well as drug administration. Additionally, the lactation stage appears to play a significant role in the breast milk microbiota, thus affecting the intestinal microbiome composition of infants, too [89,90]. Finally, antibiotics may delay intestinal colonization of potentially beneficial bacteria and reduce the diversity of the microbiome, thus predisposing to NEC [84]. For these reasons, the investigation of the gut microbiota through culture-independent techniques has increased steadily in the last years. Ideally, the use of both culture-based and culture-independent approaches should be complementary, as the first allows the isolation of bacteria at low levels in samples and even when undetectable by quantitative PCR (qPCR), while the latter enables the identification of uncultivable bacteria [35,41]. The most common molecular untargeted strategy, the 16S rRNA amplicon sequencing and community profiling, is based on the amplification of the V2-V4 regions. This approach has the advantages of generating the lowest error rate when assigning taxonomy, and of providing more reliable richness estimation when converting longer reads of the V3-V4 regions [47]. Nevertheless, the drawback of culture-independent data is the inability to resolve strains with "pathogenic potential" from non-pathogenic members of a lineage [55], as well as the reliability on the PCR method without the potential of detecting plasmids, eukariotes, and viruses, like the genome-resolved methods do [48].
In our review, four studies suggested a predominance of Proteobacteria [43,45,53,54] and two studies a reduced abundance of Firmicutes [43,54] in cases near disease onset. Two studies [26,45], instead, reported a reverse trend in the first week of life, with a dramatic shift related to these two phyla over the week before NEC diagnosis (around 34-50% increase in Proteobacteria in NEC cases). These findings suggest a change in the gut microbiota occurring after birth and progressing towards NEC development. Preterm infants do have a dynamic pattern of early intestinal colonization [81,85], which in NEC patients seems to be characterized by gram-positive cocci at the beginning, then overtaken by gram-negative facultative anaerobic organisms, counterbalanced by a gradually increasing abundance of anaerobes [29,36,47,50,81]. This evidence has also been supported by Pammi et al. [91], who conducted a rigorous review and meta-analysis of eight studies applying stool microbiome profiling and revealed a consistent trend towards higher relative abundances of Proteobacteria and decreased relative abundances of Firmicutes and Bacteroidetes around 30 weeks of corrected GA in NEC patients. Additionally, studies exploring the microbiome in intestinal specimens showed that Proteobacteria were the most abundant phyla in NEC infants (49.0%) [66], and significantly higher in these patients compared to those affected by SIP [67]. An opposite trend seems to occur in healthy controls, with a progressive shift towards more gram-positive bacteria (Firmicutes and Bacteroidetes phyla, genera Staphylococci and lactate-producing bacilli), converging to an "adult-like" microbiota phenotype [18,92,93].
Despite the discrete uniformity of findings at the phyla level, data on bacteria abundances at class, genus, and species levels showed a wider heterogeneity of results. Among gram-positive bacteria, the genera Enterococcus and Veillonella (belonging to Firmicutes) [26,43,44,46,47,51,57], and Propionibacterium and Bifidobacterium (belonging to Actinobacteria) [29,41,53] were more abundant in healthy controls than in affected subjects prior to NEC. These strains have indeed been described as beneficial for a healthy intestinal development and function [94]. Additionally, in a recent review and network meta-analysis Van den Akker et al. [95] demonstrated the superiority of combinations of one or more Bifidobacterium spp. and one or more Lactobacillus spp. in probiotics' composition in reducing the risk of severe NEC. The combination of Enterococcus spp, Bacillus spp, one or more Bifidobacterium spp and Streptococcus salivarius seemed to produce the largest reduction in NEC onset [96][97][98]. Among gram-negative bacteria, from our review three species emerged to be prevalent in NEC patients and all belonging to Proteobacteria phylum, in particular Bacteroides dorei [40], Morganella morganii [42] and a strain matching closest to Klebsiella pneumoniae [53]. The first is a known intestinal commensal strain which has been demonstrated to be involved in autoimmune diseases, like Type 1 diabetes. The latter two, instead, are renowned causes of nosocomial infections. The recent large-scale gut microbiome meta-analysis performed by Tarracchini et al. in 2021 [82] has also shown an increase in relative abundance of opportunistic pathogens in NEC, including members of the Klebsiella genus. Gramnegative bacteria are rich in lipopolysaccharide, which binds to toll-like receptor 4 (TLR-4), known to be involved in the inflammatory response of NEC [76]. The role of other gramnegative bacteria, like Pseudomonas and Escherichia coli genera, is still controversial in the included studies.
As regards anaerobes, there are several studies applying culture and non-culture methods that support the association of the genera Clostridium and Clostridium sensu strictu [49,57] with NEC, in particular with a predominance of the species Cl. Butyricum [35,41,49] and its closerelated Cl. Neonatale [49,99]. The first species was first described in NEC several decades ago [100], whereas the latter was associated with a NEC outbreak in a NICU [101]. The more well-known Clostridium perfringens and perfringens-like strains were found in faecal samples both early in life and near NEC diagnosis in two studies in our review [40,41]. These species have been associated to NEC development, especially with a fulminant course, in previous researches and case reports [102][103][104]. Similarly, Tarracchini et al. found a predominance of C. perfrigens and C. neonatale in NEC compared to healthy preterms in the analysis of only pre-NEC samples [82]. These findings suggest that some early microbial signatures involving nosocomial pathogens, which may likely be a consequence of broad-spectrum antibiotics treatment, should be further investigated.
In our review, the majority of studies found no difference in the Chao-i [26,39,45,53], but a cauterization of samples according to case status from several weeks before NEC up to NEC diagnosis by UniFrac metrics [26,44,46,53,54]. As suggested by Torrazza et al. [53], this may indicate that patients with NEC may have a total number of bacteria similar to that of healthy controls, but a difference in the kinds of OTUs present and their proportions. While the Simpson index was also comparable between cases and controls, similar to the Chaoi [26,40,43], the Shannon diversity index, another measure accounting for both abundance and evenness of the species present in a community, was lower in NEC cases in six out of eight studies [35,44,46,54,56,57]. A limited Shannon diversity usually characterizes all preterm infants, with an average of less than 20 OTUs in contrast to more than 200 for human adults [93]. However, a much more reduced index in NEC patients may indicate a decreased uniformity of individual distribution among species, and not just a lower number of species. Furthermore, given the non-complex gut microbial populations of preterm infants, the overrepresentation or underrepresentation of single taxons make it difficult to ascribe a host phenotype to a variation in diversity [105]. Pammi et al. [91] reported no differences in the observed species, Shannon diversity, and Simpson diversity indices, between NEC cases and controls, nor did the authors found any clustering of samples based on weighted and unweighted UniFrac distance. The shotgun metagenomics data collection by Tarracchini et al. [82], instead, showed that cases with manifested NEC (64 samples) were characterized by a lower index of bacterial species richness on average than that of age-matched healthy preterms. All in all, a statistically significant reduction in biodiversity in NEC compared to preterm healthy controls according to gestational age and postnatal age still needs further evaluation.
We explored studies applying both metabolomic and microbiota analysis to illustrate potential connections of the human gut according to disease status. Morrow and colleagues [26] found an association between individual urine amino acids and the microbiome preceding NEC development. In this study, alanine, which is ubiquitously incorporated into bacterial cell wall biosynthesis, especially into peptidoglycan of most gram-positive organisms, was increased in NEC preceded by Firmicutes and inversely correlated with the relative abundance of Proteobacteria and Propionibacterium. The study of the metabolome-microbiome interplay [106,107], despite being increasingly popular, needs further investigation. Studies, like the MAGPIE [108], will hopefully shed a light on this linkage [71]. Studies investigating preclinical alterations in fecal volatile organic compounds (VOCs) also harbor great potential for this purpose, as these chemicals are considered to reflect gut microbiota composition and concurrent metabolic activity in the host [74]. Indeed, several studies have demonstrated that VOC profiles of infants with NEC can be discriminated from those of controls from 2 to 4 days before onset of symptoms [109][110][111].
Our review has different limitations. Firstly, as highlighted by many previous reviews, the majority of included studies comprises a small sample size, with variable inclusion criteria, uncontrolled confounding biases and different applied techniques, therefore hindering consistent comparison of data [10]. Secondly, none of the included studies compared NEC versus SIP, which is the most common surgical disease in the differential diagnosis of NEC. These two entities share a similar clinical presentation, although with some different features. Therefore, it would be a major interest to compare SIP vs. NEC cases at the time of diagnosis in order to highlight the different pathogenetic mechanisms and guide therapeutic management. Another limitation of our study is the lack of accurate and detailed metadata analysis due to missing exact quantitative parameters, such as peak values, sensitivity, specificity, predictive values, and areas under the curve from most of the studies. These quantitative data are often not specified in the methods and results sections. Comprehensible and comparable information for future science would be possible if studies reported exact mean values with error indicators for every significant parameter, as well as the degree of significance of elevation in the comparison of study groups. Finally, caution must be paid in data interpretation due to the numerous host variables that may affect the host's metabolome and gut microbiota. In fact, as NEC progression usually occurs over 8-9 weeks in extremely preterm infants, several factors are implied in disease onset [28]. Given this consideration, the ideal time to predict NEC development would be the first week of life. However, samples collected at this time may not be as diriment as those collected closer to NEC development [10,60]. We suggest, whenever possible, to non-invasively collect the majority of samples, from birth onwards, as well as at suspicion of disease onset, in order to capture the metabolic and microbiome changes through time.
Although we did not perform a meta-analysis based on raw 16S rRNA gene sequence data, we believe this article clearly summarizes the current evidence and provides potential inspiration points for future research. In Figure 3, we provide a possible flowchart for future studies with the aim of improving uniformity of data collection and quantitative measures. Infants with SIP should be included as this entity goes in differential diagnosis with NEC, has a steady prevalence, and is a major reason for peritoneal drainage and abdominal surgery in ELBWI [112]. Validation cohorts with targeted approach and culturebased analysis are mandatory to corroborate and confirm previous "high-risk" putative biomarkers. It is probable that a sequential collection of samples rather than only one sample collected at a single timepoint would be relevant in the understanding of the hostpathogen interactions in NEC development. Algorithm comprising maternal and neonatal variables (clinical or radiological) and characteristic metabolic-microbiota patterns may be the key to early unravel surgical NEC [113][114][115].

Conclusions
Untargeted metabolomic and gut microbiota analyses applied to non-invasive specimens are valuable tools to understand NEC pathophysiology and identify its early biomarkers. These strategies are dampened by the influence of multiple confounding factors, and by the limited data on GA and age-specific reference values. The inclusion of bigger samples sizes, of patients affected by NEC or SIP as strictly defined by major guidelines, and of quantitative measures of results, is mandatory for future studies in order to improve uniformity of data. Data sharing via biorepositories is fundamental to guarantee consistent comparison of data. Future studies need to include validation cohorts to confirm the results.