AbstractAlterations in fecal microbiota and volatile organic compound (VOC) profiles of preterm infants have been demonstrated before onset of necrotizing enterocolitis (NEC). However, NEC-specific signatures need to be identified before potential application as predictive biomarker in clinical practice. A prospective multicenter case–control study was conducted to identify preclinical fecal microbiota and VOC profiles of infants that developed NEC. Microbiota analysis (PCR-based IS-pro technique) and VOC analysis (gas chromatography-mass spectrometry) were performed on fecal samples collected up to three days before clinical NEC onset. In 112 infants (56 NEC, 56 matched controls), sufficient number fecal samples were collected for either microbiota or VOC analysis. Prior to NEC onset, Clostridium perfringens (p = 0.023, unadjusted) was more present in infants with NEC, versus controls. VOC analysis showed a clear distinction between fecal profiles of NEC cases and controls (area under the curve = 0.82). Fourteen unique VOC features contributed to this discrimination. Fecal microbiota and VOC profiles may serve as early indicators of NEC, and allow for increased understanding of pathophysiological mechanisms of NEC, but larger validation cohorts are needed before an overarching NEC-specific predictive microbiota-based biomarker can be implemented.
IntroductionNecrotizing enterocolitis (NEC) is a devastating gastrointestinal disease predominantly affecting preterm infants and is associated with mortality rates of up to 30% in advanced stages1. The pathogenesis of NEC is considered multifactorial, including gut immaturity, microbial dysbiosis, and intestinal ischemia by hypoperfusion, resulting in an exaggerated immune response of the gut2. However, the exact pathogenesis of NEC remains to be elucidated, limiting the development and implementation of targeted diagnostic and therapeutic strategies. In clinical practice, NEC diagnosis is established on a combination of clinical symptoms, laboratory markers of inflammation, and radiographic findings. However, these parameters lack specificity and typically manifest only in advanced stages of the disease3,4,5. Early recognition of NEC remains challenging and may delay medical intervention, and as a consequent negatively impact clinical outcomes5.The consistent observations of microbial dysbiosis preceding NEC propose the potential for microbes to serve as a predictive biomarker6,7,8,9,10,11. However, no NEC-specific microbial signature has yet been identified, largely due to the limited number of inclusions and the lack of standardization of the study design, including the applied microbiota analysis technologies, sampling conditions, and analyzed time-points, limiting the comparison of outcomes12. In addition, fecal volatile organic compound (VOC) analysis has been postulated as a potential predictive biomarker for different neonatal morbidities including NEC13,14,15,16,17. Fecal VOCs are produced by metabolic processes of the host and gut microbiota and by their mutual interaction, and are therefore suggested to reflect both microbiota composition and their function18. However, similarly to microbiota studies, a NEC-specific VOC signature has not yet been identified. This is because the majority of studies utilized eNose technology to describe VOC profiles, which relies on profile recognition rather than the identification of unique chemical molecules. In this study, we aimed to identify microbial and VOC signatures at the molecular level preceding NEC onset, and hypothesized that the preclinical fecal microbiota and VOCs can discriminate infants with NEC from matched controls.ResultsBaseline characteristicsA total of 1002 infants were included during the study period, of whom eighty-three developed NEC (8.3%). Supplementary Table 1 shows the number of infants included in the study per participating center. From 56 infants with NEC, sufficient fecal samples and mass was collected for further analysis. Clinical and demographic data are provided in Table 1. In Supplementary Table 2, the number of analyzed fecal samples per analytical technique and time point is outlined.Table 1 Demographics.Full size tableMicrobiota analysisThere were 53 infants with NEC (NEC IIA, n = 17; NEC IIB, n = 17; NEC IIIA, n = 8; NEC IIIB, n = 11) matched to 53 controls for microbiota analysis. In total, 240 case and control samples were analyzed (Supplementary Table 2). Cumulative IS-pro microbiota profiles of all stage NEC case (n = 120) and control (n = 120) samples were constructed, which demonstrated that Clostridium perfringens (p = 0.023, adjusted p = 0.15) was significantly more present in infants that developed NEC. C. perfringens was present in respectively 26% of infants that developed NEC, compared to 9% of the control infants (Fig. 1). Phylum diversity and abundances were assessed between all NEC case (n = 120) and control samples (n = 120) at the three predefined time points. Diversity indices, including all phyla combined and Bacteroidetes, FAFV, and Proteobacteria separate, did not differ between cases and controls (Fig. 2a; Supplementary Table 3). No differences were observed in absolute and relative abundances of cases versus controls for each individual phylum and for all phyla combined (Fig. 3a, b). In addition, for both diversity indices and the absolute and relative abundance, no clear time-associated trend from t-3 to t-1 was observed (Figs. 2b, 3c).Fig. 1Cumulative IS-profiles of infants with NEC (all NEC-stages) versus controls. The peak length is displayed on the x-axis and corresponds to IS-fragment length. The peak height on the y-axis reflects the number of infants in which the corresponding peak was found. (A) In infants with NEC, Clostridium perfringens (peak length 235; p = 0.023, FDR adjusted p = 0.15) was significantly more observed compared to controls. (B) The highlighted version of A. IS, interspace; NEC, necrotizing enterocolitis; FAFV, Firmicutes, Actinobacteria, Fusobacteria and Verrucomicrobia.Full size imageFig. 2Shannon diversity index for all phyla between NEC cases and controls. (A) The Shannon diversity index is displayed for all phyla combined, and the three phyla (Bacteroidetes, FAFV, and Proteobacteria) separate. The image shows the differences between all NEC cases versus control group for all three time points separately. There were no significant differences observed. (B) The difference in Shannon diversity index is displayed over time from three days before diagnosis (t-3) to one day before diagnosis (t-1) for both NEC cases and controls for all phyla combined, and the three phyla separately. Each point resembles a sample. A line is drawn between samples from the same infant. There were no significant differences in Shannon diversity over time. NEC, necrotizing enterocolitis; FAFV, Firmicutes, Actinobacteria, Fusobacteria and Verrucomicrobia; − 3, − 2, and − 1 represent three, two and one day(s) before diagnosis, respectively.Full size imageFig. 3Mean absolute and relative abundance for all phyla per time point between NEC cases and controls. (A) The mean absolute abundance per time point (− 3, − 2, and − 1), phylum (Bacteroidetes, FAFV, and Proteobacteria), and study group (NEC vs controls) is displayed. There were no significant differences. (B) This shows relative abundance per time point, phylum and study group. (C) The mean absolute abundance is displayed over time from three days before diagnosis (t-3) to one day before diagnosis (t-1) for both NEC cases and controls for the three phyla separately. Each point resembles a sample. A line is drawn between samples from the same infant. There were no significant differences in mean and absolute abundance, and no differences in mean absolute abundance over time. NEC, necrotizing enterocolitis; FAFV, Firmicutes, Actinobacteria, Fusobacteria and Verrucomicrobia; − 3, − 2, and − 1 represent three, two and one day(s) before diagnosis, respectively; RFU, relative fluorescence units.Full size imageSubsequently, NEC cases were categorized into subgroups based on NEC staging, and only the last sample obtained before diagnosis was included in the analysis (Tlast). It was demonstrated that NEC IIIB cases (perforated gut) had a different microbial composition and diversity compared to their matched controls at Tlast (11 case samples vs. 11 control samples) (Supplementary Table 4). The observed discrimination between NEC IIIB and controls was driven by decreased overall α-diversity (p = 0.026) and a lower absolute abundance of FAFV (p = 0.041) and Bacteroidetes (p = 0.018) in cases.VOC analysesFor GC–MS analysis, 44 infants with NEC (16 NEC IIA, 12 NEC IIB, 7 NEC IIIA, and 9 NEC IIIB) were matched with 44 controls. In total, 208 samples were analyzed, including 104 NEC samples and 104 control samples. Three control samples were excluded for further analysis due to unreliable measurements (Supplementary Table 2). The number of unique VOCs isolated from the headspace of the fecal samples ranged from 101 to 382 per sample. The majority of these VOCs belonged to aldehyde and alcohol subgroups (Supplementary Table 5). Examples of chromatograms at three predefined time points are displayed for one control infant and one NEC case in Supplementary Figure 1.Similarity analysis of fecal VOC profilesFirst, the similarities between all sample pairs in the VOC profiles were assessed (Fig. 4). It was shown that all VOC profiles were moderately comparable to each other, with an overall cosine similarity peak of 0.7 (Fig. 4a). The similarity index between the samples increased to 0.9 when the matched case–control groups were taken into account independently, indicating that case samples were more similar to each other, and control samples were more similar to each other (Fig. 4b). Regarding intra-individual similarity, VOC profiles seemed to be more uniform over time in controls than in NEC cases (Fig. 4c).Fig. 4Similarities between samples as measured by cosine similarity measure. Similarity values range between zero and one. Higher cosine similarity values demonstrate an increase in sample similarity. (A) Includes all sample pairs irrelevant of disease state. This demonstrates that all samples are moderately similar (similarity index 0.7). (B) Shows the similarity between study groups, indicating that samples within the NEC and within the control group are more similar to each other (similarity index 0.9). (C) Demonstrating intra-individual similarity. This shows that control samples are more stable over time compared to NEC samples. NEC, necrotizing enterocolitis.Full size imageFecal VOC profile analysisSecond, supervised classification models were applied to discriminate the fecal VOC patterns of NEC samples (n = 104) from control samples (n = 101), resulting in a moderate to low diagnostic accuracy. The best test accuracy was obtained when combining VOC profiles of all three time points (area under curve (AUC) [95% CI]: 0.71 [0.63–0.78]) using GLMVQ (Supplementary Figure 2, Supplementary Table 6).Single-day and longitudinal fragment and chromatogram feature analysisThe longitudinal course of unique compounds and chromatogram characteristics of the VOC profiles before NEC diagnosis were assessed by analyses including case and control infants with samples analyzed at all three pre-defined timepoints (n = 24 infants with NEC, n = 23 control infants, n = 141 samples). The analysis was performed based on single-day or day-difference fragment and chromatogram features, or a combination of these features, as displayed in Supplementary Table 7 and Table 2. Three classification algorithms were applied to obtain the best performing feature set. The best classification performance was achieved using the logistic regression (LR) classifier by combining all feature sets (FFC) (accuracy = 77.7% and AUC = 0.82) (Table 2, Fig. 5a). In Fig. 5b it is shown that optimal accuracy when combining all feature sets (FFC) was obtained using 14 unique features (AUC = 0.82). The fourteen features that were selected most frequently by LR are displayed in Table 3, of which eleven were significantly different (p < 0.05) between NEC samples and controls (Supplemental Figure 3). Eight of these features were from samples obtained 2 days before diagnosis of NEC (Table 3). Furthermore, eight of these single-day features were identified as unique chemical compounds, namely 3-(methylthio)-propanal (p = 0.003, t-1), camphene (p = 0.017 and p = 0.015, t-2), benzene acetaldehyde (p = 0.006, t-2), 3-methylbenzaldehyde (p = 0.006, t-2), tert.butoxy-benzene (p = 0.045, t-2), 2-ethyl-1,3-butadiene or cyclohexene (p = 0.067, t-3), and 2-pentylfuran (p = 0.025, t-1), while one feature was unidentified (p = 0.006, t-2), (Table 3, Supplementary Figure 3).Table 2 Performance of classification model on different feature sets by logistic regression.Full size tableFig. 5Receiver-operating-characteristic curves and features performance by logistic regression classification. (A) Receiver-operating-characteristic curves for the different feature sets as obtained by logistic regression classification (n = 44 infants with NEC vs. 44 controls). Diagonal dashed line represents a random guess. (B) Influence of the number of features on the performance of the logistic regression classification model. A plateau is reached when 14 features are included in the classification model. Including more than 15 features did not result in an improved diagnostic accuracy in differentiating infants with NEC from controls (n = 44 infants with NEC vs. 44 controls). AUC, area under curve; Fs-frag, fragment feature including single day features (e.g. one of the three time points); Fd-frag fragment feature differences between two time points (e.g. between day one and two or day two and three); Fs-chro, chromatogram feature including single day features; Fd-chro, chromatogram feature differences between two time points; Ffrag fragment feature in which all time points and differences between time points are included; Fchro, chromatogram feature in which all time points and differences between time points are included. FFC, all fragment and chromatogram features were included in the classification model.Full size imageTable 3 The fourteen features included to obtain most optimal classification model using fivefold cross validation with logistic regression.Full size tableDiscussionIn this multicenter case–control study, we aimed to identify fecal microbial and volatile metabolic signatures in the three days preceding NEC onset. C. perfringens was significantly more abundant in infants with NEC compared to controls. Microbiota composition differed significantly from controls in NEC IIIB cases, which was driven by a decreased overall α-diversity, lower absolute abundance of FAFV, and depletion of Bacteroidetes. Fourteen VOC features, of which eight unique VOCs differed significantly between cases and controls prior to diagnosis, however, most of these compounds differed only two days prior to clinical onset. Based on the single- and day-difference fragments and chromatogram features combined, NEC could be distinguished from controls with an AUC of 0.82.Several microbial strains have been proposed to be associated with development of NEC. Here, we demonstrated that C. perfringens was significantly more abundant prior to clinical NEC onset compared to controls, with an unadjusted p value of p = 0.023. Clostridia species have consistently demonstrated involvement in the development of NEC11,19,20,21,22,23,24,25,26. NEC associated with C. perfringens, as found in the peritoneal cavity, blood, or stool, is more severe and comprises higher mortality and morbidity24. Similar to the current study, a higher abundance of C. perfringens in pre-NEC samples was demonstrated in two studies23,26. One study showed an increased in C. perfringens from birth up to NEC diagnosis (11 NEC cases and 22 matched controls)23. Another study identified three pre-NEC taxonomic profiles which were all dominated by well-recognized (opportunistic) pathogens26. One of these pre-NEC community types was dominated by C. perfringens, amongst other pathogens, with a significant higher abundance of C. perfringens in pre-NEC samples versus healthy controls (26.52% and 1.02%, respectively)26. Previous research showed that NEC-specific clostridial strains can produce interleukin-8 in Caco-2 cells, suggesting they might induce inflammatory processes involved in the pathogenesis of NEC27.We did not observe differences in α-diversity before NEC onset, which is consistent with results from a meta-analysis including 8 studies with 106 NEC and 278 controls11. However, a systematic review including 20 studies with 254 NEC cases and 673 controls reported a decrease in α-diversity10, and Tarracchini et al. reported lower biomass in pre-NEC samples26. In these studies, a trend was demonstrated towards an increased relative abundance of Proteobacteria10,11,26, and decreased levels of Firmicutes and Bacteroidetes in NEC cases10,11. We observed a decreased absolute abundance of FAFV and absence of Bacteroidetes in infants with NEC IIIB. The observed microbial differences between the different stages of NEC may suggest the presence of phenotype-specific metabolic and microbial pathways.In the current study the fecal VOC profiles of NEC cases could be distinguished from controls with an AUC ranging from 0.6–0.8, by applying supervised classification models (Supplemental Figure 2). Analyzing the single-day and day-difference fragment and chromatogram features revealed that eight unique compounds belonging to aldehydes and alcohols, strongly contributed to the observed discrimination between NEC and controls. Comparable to our results, Probert et al. (32 NEC cases vs. 70 controls) showed several unique compounds belonging to aldehydes and alcohols resulted in the differentiation of NEC from controls with AUCs of 0.75–0.76 up to four days prior to clinical onset14. One of these VOCs, 2-pentylfuran, also contributed to the discrimination of cases versus controls in the current study. Contrary, Garner et al. (6 NEC cases vs. 7 controls) demonstrated differences in the group of esters in fecal samples of infants with NEC, up to four days prior to clinical onset15.One of the discriminative aldehydes detected in this study, 3-methylthiopropanal, is synthesized by specific bacteria belonging to Firmicutes, including Lactococcus lactis28. This compound is a prostaglandin antagonist that possibly mitigates the inflammatory effects of prostaglandins29. We demonstrated that a reduced absolute abundance of FAFV is associated with the development of NEC IIIB, and thus hypothesize that lower 3-methylthiopropanal levels may result from the reduced abundance of Firmicutes. Other aldehydes, including benzene-acetaldehyde and 3-methylbenzaldehyde, have been associated with the development of Crohn’s disease30. It has been hypothesized that aldehydes are products of lipid peroxidation, which is thought to occur due to epithelial inflammation30. The presence of these compounds could serve as early markers of enterocolitis. Further studies are required to confirm these hypotheses.Based on an extensive literature and mVOC 4.0 (Microbial Volatile Organic Compounds) database search, the observed set of 14 VOC features could not specifically be linked to C. perfringens31. It is important to note that fecal VOCs, in addition to originate from metabolic activity of microbes, also have been described to originate from host metabolism and host-environmental factors, but the precise contribution of each factor on VOC composition remains incompletely understood32. These multiple VOC sources complicates the interpretation of findings in VOC studies and should be taken into account in future research and clinical applications.Although we demonstrate differences in VOC features in fecal NEC samples versus controls, our findings on VOCs as predictive marker for NEC are not completely consistent with findings of earlier studies13,14,15, hampering validation for use as tool in daily clinical practice. Interhospital differences influencing fecal VOC composition could play a role, but variations in sampling and analysis methods between studies need to be considered as possible explanation for the apparent differences in VOC outcomes as well33,34. Therefore, uniform, publicly available standard operating procedures (SOPs) should be applied in future studies, allowing for better inter-study comparisons. Such a proposal has been described in a previous study in which optimal sampling conditions for fecal VOC analysis by GC–MS were studied and applied in the current study35. Next to the need for a SOP, the well-known difficulties in the diagnosis and staging of NEC may also complicate classification of NEC in studies36,37. NEC has a multifaceted clinical presentation and typical signs of NEC such as pneumatosis are often not overtly present in the most preterm infants. Therefore, despite our best efforts, there is a risk of misclassification of NEC38,39, particularly in the lower stage cases with no surgical and pathologic confirmation. Even a sporadic misclassification of NEC may negatively impact the predictive value of NEC biomarkers. The higher accuracy in the NEC IIIB cases may, next to the pathophysiological processes, possibly be result from more certain classification of these specific cases.The major strength of this study is its prospective multicenter study design, allowing for the inclusion of a large number of NEC cases, enabling the assessment of different techniques for their potential as non-invasive preclinical diagnostic biomarkers. Furthermore, a uniform collection method was applied for all infants regarding clinical data and sample collection, the latter minimizing the risk of bias by sampling variation. Moreover, the large dataset allowed for a strict matching procedure based on gestational age and center of birth, limiting these two factors as possible confounders40,41. VOC analysis was done according to an optimized sampling protocol35. Finally, we applied a novel PCR-based microbiota analytical technique that enables high-throughput analyses and with the potential to generate outcomes at the species level within five hours after sampling42. Consequently, clinicians may utilize this technique in their clinical decision-making process.One of the main limitations was the limited fecal sample mass available per included infant. Hence, simultaneous analyses of microbiota and VOCs on all fecal samples was not feasible. It was therefore not possible to combine VOC and microbiota outcomes, limiting the insight into the functional roles of the bacteria in the development of NEC. In addition, the observed difference in the presence of C. perfringens between infants with NEC and controls did not remain significant after correcting for multiple testing. However, this study is explorative and hypothesis-generating, aimed at identifying potential microbes associated with NEC. Given that Clostridia spp. have previously been linked to NEC, we chose to report these results. We acknowledge that future studies are essential to further validate this microbial signature and confirm its association with NEC.In conclusion, we demonstrated differences in fecal microbiota and VOC composition up to 3 days before NEC onset in cases versus matched controls. Our findings underline the potential of fecal microbiota and VOC profiles as early diagnostic biomarkers of NEC, but illustrate that future studies with uniform SOPs for collection and analysis and flawless classification of NEC, by only including the more severe stages of NEC, are needed to validate the current results. Ideally, future studies should use a multi-omics approach to integrate the microbiota composition and fecal VOC profiles, gaining more insight into the gut microbiota function in NEC pathophysiology. This could aid in the development of a fecal microbiota-based non-invasive biomarker for identification of infants at risk of NEC that can be used in daily clinical practice.MethodsSubjectsIn this prospective case–control study infants born ≤ 30 weeks of gestation in nine participating neonatal intensive care units (NICUs) in the Netherlands and Belgium (Supplementary Table 1) were eligible to participate13,43. Infants that developed NEC (Bell’s stage ≥ IIA) in the first 28 days of life were included in the study44. All NEC cases were independently reviewed and staged based on clinical, radiographic, laboratory, and, if present, pathological data by two expert clinicians (HN and TM), in which consensus was reached in all cases. Case and control infants were excluded in case of major congenital gastrointestinal diseases, isolated spontaneous intestinal perforation (SIP), early-onset sepsis (EOS), abdominal surgery unrelated to NEC, or insufficient fecal sample mass (< 75 mg per sample). Late-onset sepsis (LOS) was an additional exclusion criteria, however, infants with NEC that developed LOS within 3 days before or during NEC onset were not excluded. To avoid duplicate publication, infants were excluded for VOC analysis if previously analyzed13.Sample analysisSample and data collectionFecal samples were collected from the diaper, daily from birth up to 28 days postnatally, transferred into a stool container (Stuhlgefäß 10 mL, Frickenhausen, Germany), and stored at − 20 °C until further handling. Probiotics were not administered during the sampling period across participating centers. For analysis, fecal samples from infants with NEC were selected one, two, and three days (t-1, t-2, and t-3, respectively) prior to clinical onset (t0). Infants with NEC were matched to controls based on gestational age (± 2 days), center of birth, and sample availability at postnatally age-matched days (e.g. every sample from an infant with NEC was postnatally age-matched to a control sample). In case of insufficient fecal sample mass (< 150 mg per sample) to perform both analyses, the case–control sample pairs were randomly selected for microbiota or VOC analysis.Clinical and demographic data were prospectively collected from patient records (Table 1). Feeding practices were categorized into three subgroups: (1) human milk (all days before t0 consisted of a daily intake of > 75% human milk) (2) formula milk (all days before t0 consisted of a daily intake of > 75% formula milk) , and (3) a combination of human milk and formula milk (the days before t0 varied between days with > 75% human milk, > 75% formula milk, or 25–75% human milk supplemented with formula milk).Microbiota analysesSample preparation, DNA extraction and microbiota analysisA detailed description of the applied IS-pro technique, a molecular microbiota profiling method with capacities to generate data on microbiota composition at the species level, generating outcome within several hours following sampling, has been published previously42,45. IS-pro analyzes the entire microbial community in a sample without targeting specific organisms. A summary of sample preparation, DNA extraction, and microbiota analysis is provided in Appendix A. In short, three different fluorescent-labelled PCR primers were used to amplify the phyla Bacteroidetes and Proteobacteria separately, and Firmicutes, Actinobacteria, Fusobacteria, and Verrucomicrobia (FAFV) combined. The interspacer length of the 16S-23S rDNA region, expressed as the number of nucleotides, was used to identify the bacteria at the species level.Fecal VOC analysesSample preparation and analysisFecal VOC analysis was performed using gas chromatography–time-of-flight mass spectrometry (GC-ToF–MS) (Agilent 7890 B GC-TOF, LECO Pegasus 4D system, LECO, St. Joseph, MI, USA). A detailed description of the sample preparation and device settings is provided in Appendix B and is in accordance with a previously published study35. The software used for the operation of the equipment and data analysis included LECO Chromatof (LECO, St. Joseph, MI, USA) and the National Institute of Standards and Technology (NIST) mass spectral library version 2.0 build 2012 (NIST, Gaithersburg, MD, USA).Data preprocessingThe measured chromatograms were processed according to the method described by Smith et al., as was the execution of peak detection and alignment46. To compensate for the small drifts in the GC–MS output, the chromatograms were normalized using hexafluoroisopropanol as the internal standard (ISTD). Because the ISTD can be influenced by the VOC content in the fecal samples, the ISTD in the quality control samples was used for normalization at the start and end of each batch.Statistical analysesClinical and demographical dataSPSS® (version 26.0, IBM, NY) was utilized for statistical analyses of clinical and demographic data. Independent samples t-test, Chi-squared test, or Mann–Whitney U test were applied when appropriate. Data distribution was assessed through histogram shape, employing the Shapiro–Wilk test in case of uncertainty.Microbiota analysesStandard IS-pro proprietary software (Fragment Analyser V0.22.3) was used for the statistical analysis which is elaborated on in Appendix C47. In short, α-diversity was determined using the Shannon-diversity index, differences in absolute and relative abundance were assessed using the Mann–Whitney U-test, and intra-individual progression was evaluated by comparing abundance and diversity from t-3 to t-1.VOC data analysesThe data analysis regarding VOC analysis is described in detail in Appendix C. The similarity analysis and supervised classification were performed using R (version 3.3.3) with the R–studio interface48. In short, the similarity between the samples was estimated by calculating the cosine similarity measure. To discriminate NEC VOC profiles from controls, the supervised classification methods sparse partial least squares discriminant analysis (sPLS-DA), least absolute shrinkage and selection operator (LASSO), and generalized matrix learning vector quantization (GLMVQ) using tenfold cross validation were applied49,50,51,52.Next, infants with samples collected at all three predefined time points were included in an additional analysis to include fragment and chromatogram differences over time (e.g. difference between t-1 and t-2), as well as single-day (e.g. t-1, t-2, or t-3) chromatogram and fragment features (Supplementary Table 7). Different types of features sets were established for these analyses (Appendix C). Optimal classification models were constructed using these selected feature sets, including logistic regression, k-nearest neighbors, and support vector machines.
Data availability
The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to containing information that could compromise the privacy of research participants.
ReferencesThyoka, M. et al. Advanced necrotizing enterocolitis part 1: Mortality. Eur. J. Pediatric Surg. 22(1), 8–12 (2012).CAS
Google Scholar
Eaton, S., Rees, C. M. & Hall, N. J. Current research on the epidemiology, pathogenesis, and management of necrotizing enterocolitis. Neonatology 111(4), 423–430 (2017).PubMed
MATH
Google Scholar
Rich, B. S. & Dolgin, S. E. Necrotizing enterocolitis. Pediatr. Rev. 38(12), 552–559 (2017).PubMed
Google Scholar
Sharma, R. & Hudak, M. L. A clinical perspective of necrotizing enterocolitis: Past, present, and future. Clin. Perinatol. 40(1), 27–51 (2013).PubMed
PubMed Central
MATH
Google Scholar
Neu, J. Prevention of necrotizing enterocolitis. Clin. Perinatol. 49(1), 195–206 (2022).PubMed
MATH
Google Scholar
Fu, X., Li, S., Jiang, Y., Hu, X. & Wu, H. Necrotizing enterocolitis and intestinal microbiota: The timing of disease and combined effects of multiple species. Front. Pediatr. 9, 657349 (2021).PubMed
PubMed Central
Google Scholar
Liu, X. C. et al. Gut microbiota and short-chain fatty acids may be new biomarkers for predicting neonatal necrotizing enterocolitis: A pilot study. Front. Microbiol. 13, 969656 (2022).PubMed
PubMed Central
Google Scholar
Xiong, J. et al. Alterations of the gut microbiota and short chain fatty acids in necrotizing enterocolitis and food protein-induced allergic protocolitis infants: A prospective cohort study. Front. Cell. Infect. Microbiol. 12, 1030588 (2022).CAS
PubMed
PubMed Central
Google Scholar
Huang, H. et al. Abnormalities in microbial composition and function in infants with necrotizing enterocolitis: A single-center observational study. Front. Pediatr. 10, 963345 (2022).PubMed
PubMed Central
Google Scholar
Moschino, L. et al. The metabolome and the gut microbiota for the prediction of necrotizing enterocolitis and spontaneous intestinal perforation: A systematic review. Nutrients 14(18), 3859 (2022).CAS
PubMed
PubMed Central
Google Scholar
Pammi, M. et al. Intestinal dysbiosis in preterm infants preceding necrotizing enterocolitis: A systematic review and meta-analysis. Microbiome 5(1), 31 (2017).PubMed
PubMed Central
Google Scholar
Neu, J. & Pammi, M. Necrotizing enterocolitis: The intestinal microbiome, metabolome and inflammatory mediators. Semin. Fetal Neonatal. Med. 23(6), 400–405 (2018).PubMed
Google Scholar
de Meij, T. G. et al. Early detection of necrotizing enterocolitis by fecal volatile organic compounds analysis. J. Pediatr. 167(3), 562–7.e1 (2015).PubMed
MATH
Google Scholar
Probert, C. et al. Faecal volatile organic compounds in preterm babies at risk of necrotising enterocolitis: the DOVE study. Arch Dis Child Fetal Neonatal Ed. 105(5), 474–479 (2020). Garner, C. E. et al. Analysis of faecal volatile organic compounds in preterm infants who develop necrotising enterocolitis: A pilot study. J. Pediatr. Gastroenterol. Nutr. 49(5), 559–565 (2009).CAS
PubMed
Google Scholar
Hosfield, B. D. et al. The assessment of microbiome changes and fecal volatile organic compounds during experimental necrotizing enterocolitis. J. Pediatric Surg. 56(6), 1220–1225 (2021).
Google Scholar
Wright, H., Bannaga, A. S., Iriarte, R., Mahmoud, M. & Arasaradnam, R. P. Utility of volatile organic compounds as a diagnostic tool in preterm infants. Pediatric Res. 89(2), 263–268 (2021).
Google Scholar
Amann, A. et al. The human volatilome: Volatile organic compounds (VOCs) in exhaled breath, skin emanations, urine, feces and saliva. J. Breath Res. 8(3), 034001 (2014).ADS
CAS
PubMed
Google Scholar
Sim, K. et al. Dysbiosis anticipating necrotizing enterocolitis in very premature infants. Clin. Infect. Dis. 60(3), 389–397 (2015).CAS
PubMed
Google Scholar
Zhou, Y. et al. Longitudinal analysis of the premature infant intestinal microbiome prior to necrotizing enterocolitis: A case-control study. PLoS ONE 10(3), e0118632 (2015).PubMed
PubMed Central
Google Scholar
de la Cochetiere, M. F. et al. Early intestinal bacterial colonization and necrotizing enterocolitis in premature infants: The putative role of Clostridium. Pediatric Res. 56(3), 366–370 (2004).
Google Scholar
Cassir, N. et al. Clostridium butyricum strains and dysbiosis linked to necrotizing enterocolitis in preterm neonates. Clin. Infect. Dis. 61(7), 1107–1115 (2015).CAS
PubMed
Google Scholar
Heida, F. H. et al. A necrotizing enterocolitis-associated gut microbiota is present in the meconium: Results of a prospective study. Clin. Infect. Dis. 62(7), 863–870 (2016).CAS
PubMed
Google Scholar
Dittmar, E. et al. Necrotizing enterocolitis of the neonate with Clostridium perfringens: Diagnosis, clinical course, and role of alpha toxin. Eur. J. Pediatr. 167(8), 891–895 (2008).PubMed
MATH
Google Scholar
Itani, T. et al. (2018) Preterm infants with necrotising enterocolitis demonstrate an unbalanced gut microbiota. Acta Paediatrica (Oslo, Norway: 1992) 107(1), 40–47 (2018).PubMed
Google Scholar
Tarracchini, C. et al. Unraveling the microbiome of necrotizing enterocolitis: Insights in novel microbial and metabolomic biomarkers. Microbiol. Spectr. 9(2), e0117621 (2021).PubMed
Google Scholar
Schonherr-Hellec, S. et al. Clostridial strain-specific characteristics associated with necrotizing enterocolitis. Appl. Environ. Microbiol. 84(7), e02428-e2517 (2018).PubMed
PubMed Central
Google Scholar
Amárita, F., Fernández-Esplá, D., Requena, T. & Pelaez, C. Conversion of methionine to methional by Lactococcus lactis. FEMS Microbiol. Lett. 204(1), 189–195 (2001).PubMed
Google Scholar
ChEbi. 3-methylthiopropanal. Available at: https://www.ebi.ac.uk/chebi/searchId.do?chebiId=CHEBI%3A49017. Accessed 18 February.Ahmed, I., Greenwood, R., Costello Bde, L., Ratcliffe, N. M. & Probert, C. S. An investigation of fecal volatile organic metabolites in irritable bowel syndrome. PLoS ONE 8(3), e58204 (2013).ADS
PubMed
PubMed Central
Google Scholar
Kemmler, E. et al. mVOC 4.0: A database of microbial volatiles. Nucleic Acids Res. 53, D1692–D1696 (2024).PubMed Central
Google Scholar
de Kroon, R. R. et al. The potential of fecal volatile organic compound analysis for the early diagnosis of late-onset sepsis in preterm infants: A narrative review. Sensors (Basel, Switzerland) 24(10), 3162 (2024).PubMed
Google Scholar
Bosch, S. et al. Optimized sampling conditions for fecal volatile organic compound analysis by means of field asymmetric ion mobility spectrometry. Anal. Chem. 90(13), 7972–7981 (2018).CAS
PubMed
PubMed Central
Google Scholar
Berkhout, D. J. et al. Effects of sampling conditions and environmental factors on Fecal volatile organic compound analysis by an electronic nose device. Sensors (Basel, Switzerland) 16(11), 1967 (2016).ADS
PubMed
MATH
Google Scholar
El Manouni El Hassani, S. et al. Optimized sample preparation for fecal volatile organic compound analysis by gas chromatography-mass spectrometry. Metabolomics 16(10), 112 (2020).CAS
PubMed
PubMed Central
Google Scholar
Patel, R. M., Ferguson, J., McElroy, S. J., Khashu, M. & Caplan, M. S. Defining necrotizing enterocolitis: Current difficulties and future opportunities. Pediatric Res. 88(Suppl 1), 10–15 (2020).
Google Scholar
Kim, J. H., Sampath, V. & Canvasser, J. Challenges in diagnosing necrotizing enterocolitis. Pediatric Res. 88(Suppl 1), 16–20 (2020).
Google Scholar
Challis, P. et al. Factors associated with the increased incidence of necrotising enterocolitis in extremely preterm infants in Sweden between two population-based national cohorts (2004–2007 vs 2014–2016). Arch. Dis. Childhood Fetal Neonatal Ed. 109(1), 87–93 (2023).MATH
Google Scholar
Berrington, J. & Embleton, N. D. Discriminating necrotising enterocolitis and focal intestinal perforation. Arch. Dis. Childhood Fetal Neonatal Ed. 107(3), 336–339 (2022).
Google Scholar
Korpela, K. et al. Intestinal microbiota development and gestational age in preterm neonates. Sci. Rep. 8(1), 2453 (2018).ADS
PubMed
PubMed Central
Google Scholar
Rozé, J. C. et al. Assessment of neonatal intensive care unit practices and preterm Newborn gut microbiota and 2-year neurodevelopmental outcomes. JAMA Netw. Open 3(9), e2018119 (2020).PubMed
PubMed Central
Google Scholar
Budding, A. E. et al. IS-pro: High-throughput molecular fingerprinting of the intestinal microbiota. FASEB J. 24(11), 4556–4564 (2010).CAS
PubMed
MATH
Google Scholar
Berkhout, D. J. C. et al. Detection of sepsis in preterm infants by Fecal volatile organic compounds analysis: A proof of principle study. J. Pediatric Gastroenterol. Nutr. 65(3), e47–e52 (2017).CAS
Google Scholar
Bell, M. J. et al. Neonatal necrotizing enterocolitis. Therapeutic decisions based upon clinical staging. Ann. Surg. 187(1), 1–7 (1978).CAS
PubMed
PubMed Central
Google Scholar
de Meij, T. G. et al. Composition and stability of intestinal microbiota of healthy children within a Dutch population. FASEB J. 30(4), 1512–1522 (2016).PubMed
Google Scholar
Smith, C. A., Want, E. J., O’Maille, G., Abagyan, R. & Siuzdak, G. XCMS: Processing mass spectrometry data for metabolite profiling using nonlinear peak alignment, matching, and identification. Anal. Chem. 78(3), 779–787 (2006).CAS
PubMed
Google Scholar
Available at: https://antoni-research.inbiome.com/.Team RC. R: A language and environment for statistical computing. Available at: https://www.R-project.org/.Friedman, J., Hastie, T. & Tibshirani, R. Regularization paths for generalized linear models via coordinate descent. J. Stat. Softw. 33(1), 1–22 (2010).PubMed
PubMed Central
MATH
Google Scholar
Schneider, P., Biehl, M. & Hammer, B. Adaptive relevance matrices in learning vector quantization. Neural Comput. 21(12), 3532–3561 (2009).MathSciNet
PubMed
MATH
Google Scholar
Tibshirani, R. Regression shrinkage and selection via the lasso. J. R. Stat. Soc. Ser. B (Methodol.) 58, 267–288 (1996).MathSciNet
MATH
Google Scholar
Chun, H. & Keles, S. Sparse partial least squares regression for simultaneous dimension reduction and variable selection. J. R. Stat. Soc. Ser. B Stat. Methodol. 72(1), 3–25 (2010).MathSciNet
MATH
Google Scholar
Download referencesAcknowledgementsWe would like to acknowledge the efforts of all nurses in collecting fecal samples for this study. We would also like to thank Zeldzame Ziekte Fonds and Vaillant Fonds for enabling us to conduct this study by funding this research.FundingThis work was supported by unrestricted grants provided by the Zeldzame Ziekten Fonds and the Vaillant Fonds.Author informationAuthors and AffiliationsDepartment of Pediatric Gastroenterology, Emma Children’s Hospital, Amsterdam Gastroenterology Endocrinology Metabolism Research Institute, Amsterdam UMC, Meibergdreef 9, 1105 AZ, Amsterdam, The NetherlandsS. el Manouni el Hassani, N. M. Frerichs, D. J. C. Berkhout, M. A. Benninga & T. G. J. de MeijAmsterdam Reproduction and Development Research Institute, Amsterdam UMC, Amsterdam, The NetherlandsS. el Manouni el Hassani, N. M. Frerichs, M. M. van Weissenbruch & A. H. van KaamPhilips S&RC Research & Development, Eindhoven, The NetherlandsT. NijsenEurofins EAG, Eurofins Materials Science Netherlands B.V., Eindhoven, The NetherlandsH. H. KnobelNetherlands Institute for Public Safety, Arnhem, The NetherlandsH. WedaLighting and IoT Lab, Department of Electrical Engineering, Eindhoven University of Technology, Eindhoven, The NetherlandsM. XuBiomedical Diagnostics Lab, Department of Electrical Engineering, Eindhoven University of Technology, Eindhoven, The NetherlandsX. LongPhilips Research, Eindhoven, The NetherlandsL. WijnoltzDepartment of Neonatology, Emma Children’s Hospital, Amsterdam UMC, Amsterdam, The NetherlandsM. M. van Weissenbruch & A. H. van KaamNeonatal Intensive Care Unit, University Hospitals Leuven, Leuven, BelgiumV. CosseyMathematical and Statistical Methods Group (Biometris), Wageningen University & Research, Wageningen, The NetherlandsC. F. W. PeetersNeonatal Intensive Care Unit, Amalia Children’s Centre, Isala, Zwolle, The NetherlandsR. A. van LingenNeonatal Intensive Care Unit, Beatrix Children’s Hospital/University Medical Centre Groningen, Groningen, The NetherlandsC. V. HulzebosNeonatal Intensive Care Unit, Wilhelmina Children’s Hospital, University Medical Centre Utrecht, Utrecht, The NetherlandsD. C. VijlbriefNeonatal Intensive Care Unit, Radboud University Medical Centre, Radboud Institute for Health Sciences, Amalia Children’s Hospital, Nijmegen, The NetherlandsW. P. de BoodeDepartment of Pediatrics, Maastricht University Medical Center, Maastricht, The NetherlandsB. W. KramerinBiome BV, Amsterdam, The NetherlandsA. E. BuddingDepartment of Gastroenterology and Hepatology, Amsterdam Gastroenterology Endocrinology Metabolism Research Institute, Amsterdam UMC, Vrije Universiteit, Amsterdam, The NetherlandsN. K. H. de BoerNeonatal Intensive Care Unit, Máxima Medical Center, Veldhoven, The NetherlandsH. J. NiemarktAuthorsS. el Manouni el HassaniView author publicationsYou can also search for this author inPubMed Google ScholarN. M. FrerichsView author publicationsYou can also search for this author inPubMed Google ScholarD. J. C. BerkhoutView author publicationsYou can also search for this author inPubMed Google ScholarT. NijsenView author publicationsYou can also search for this author inPubMed Google ScholarH. H. KnobelView author publicationsYou can also search for this author inPubMed Google ScholarH. WedaView author publicationsYou can also search for this author inPubMed Google ScholarM. XuView author publicationsYou can also search for this author inPubMed Google ScholarX. LongView author publicationsYou can also search for this author inPubMed Google ScholarL. WijnoltzView author publicationsYou can also search for this author inPubMed Google ScholarM. M. van WeissenbruchView author publicationsYou can also search for this author inPubMed Google ScholarA. H. van KaamView author publicationsYou can also search for this author inPubMed Google ScholarV. CosseyView author publicationsYou can also search for this author inPubMed Google ScholarC. F. W. PeetersView author publicationsYou can also search for this author inPubMed Google ScholarR. A. van LingenView author publicationsYou can also search for this author inPubMed Google ScholarC. V. HulzebosView author publicationsYou can also search for this author inPubMed Google ScholarD. C. VijlbriefView author publicationsYou can also search for this author inPubMed Google ScholarW. P. de BoodeView author publicationsYou can also search for this author inPubMed Google ScholarB. W. KramerView author publicationsYou can also search for this author inPubMed Google ScholarA. E. BuddingView author publicationsYou can also search for this author inPubMed Google ScholarM. A. BenningaView author publicationsYou can also search for this author inPubMed Google ScholarN. K. H. de BoerView author publicationsYou can also search for this author inPubMed Google ScholarH. J. NiemarktView author publicationsYou can also search for this author inPubMed Google ScholarT. G. J. de MeijView author publicationsYou can also search for this author inPubMed Google ScholarContributionsD.B., M.X., and S.e.M.e.H. designed the study, collected data, and drafted the initial manuscript. T.N., H.K., H.W., H.N., T.d.M., L.W., M.v.W., A.v.K., V.C., C.P., R.v.L., C.v.H., D.c.V., W.d.B., B.K., A.B., M.B., and N.d.B. collected the data and critically reviewed the manuscript for important intellectual content. X.L. contributed to the design of the study, supervision, and critically reviewed the manuscript for important intellectual content. N.M.F. critically reviewed the manuscript for important intellectual content. T.d.M., T.N., H.K., and H.N. conceptualized and designed the study and data collection, collected data, supervised data collection, and critically reviewed the manuscript for important intellectual content. All authors approved the final manuscript as submitted and agreed to be accountable for all aspects of the work.Corresponding authorCorrespondence to
N. M. Frerichs.Ethics declarations
Competing interests
The authors declare no competing interests.
Ethics statement
The study was approved by all local Medical Ethical Review Boards (protocol number A2016.313) and written informed consent was obtained from the legal guardians of all included infants. All experiments were performed in accordance with relevant guidelines and regulations.
Additional informationPublisher’s noteSpringer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.Supplementary InformationSupplementary Information.Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.
Reprints and permissionsAbout this articleCite this articleel Manouni el Hassani, S., Frerichs, N.M., Berkhout, D.J.C. et al. Longitudinal fecal microbiota and volatile metabolomics preceding necrotizing enterocolitis in preterm infants: a case–control study.
Sci Rep 15, 10419 (2025). https://doi.org/10.1038/s41598-025-94692-wDownload citationReceived: 08 August 2024Accepted: 17 March 2025Published: 26 March 2025DOI: https://doi.org/10.1038/s41598-025-94692-wShare this articleAnyone you share the following link with will be able to read this content:Get shareable linkSorry, a shareable link is not currently available for this article.Copy to clipboard
Provided by the Springer Nature SharedIt content-sharing initiative
KeywordsMetabolomicsVolatile organic compoundsMicrobiomeStoolNecrotizing enterocolitis