新生儿粪便中挥发性有机物检测方案(感官智能分析)

智能文字提取功能测试中

JOURNAL OF SURGICAL RESEARCH·OCTOBER 2020 (254) 340-347Available online at www.sciencedirect.com 341HOSFIELDETAL·FECAL VOCSPREDICTMICROBIALENTEROTYPE https：//doi.org/10.1016/j.jss.2020.05.010 ScienceDirectjournal homepage：www.JournalofSurgicalResearch.com Association for Academic Surgery The Assessment of Fecal Volatile OrganicCompounds in Healthy Infants： Electronic NoseDevice Predicts Patient Demographics andMicrobial Enterotype Check for updates Brian D. Hosfield， MD， MS， Anthony R. Pecoraro， MD，aNielson T.Baxter， PhD， Troy B. Hawkins， PhD，and Troy A. Markel， MD，* a Department ofSurgery， Indiana University School of Medicine， Indianapolis， IndianaElanco Animal Health， Greenfield， Indiana ARTICLE IIN F O Article history： Received 9 December 2019 Received in revised form 4 May 2020 Accepted 5 May 2020 Available online xxx Keywords： Fecal volatile organic compounds Biomarkers Necrotizing enterocolitis Electronic nose Enterotype ABSTRA C T Background： The assessment of fecal volatile organic compounds (VOCs) has emerged as anoninvasive biomarker in many different pathologies. Before assessing whether VOCs canbe used to diagnose intestinal diseases， including necrotizing enterocolitis (NEC)， it isnecessary to measure the impact of variable infant demographic factors on VOC signals.Materials and methods： Stool samples were collected from term infants at four hospitals in alarge metropolitan area. Samples were heated， and fecal VOCs assessed by the Cyranose320 Electronic Nose. Twenty-eight sensors were combined into an overall smellprint andwere also assessed individually. 16s rRNA gene sequencing was used to categorize infantmicrobiomes. Smellprints were correlated to feeding type (formula versus breastmilk)， sex，hospital of birth， and microbial enterotype.Overall smellprints were assessed by PERMA-NOVA with Euclidean distances， and individual sensors from each smellprint wereassessed by Mann-Whitney U-tests. P <0.05 was significant. Results：Overall smellprints were significantly different according to diet. Individual sensorswere significantly different according to sex and hospital of birth， but overall smellprintswere not significantly different. Using a decision tree model， two individual sensors couldreliably predict microbial enterotype. Conclusions： Assessment of fecal VOCs with an electronic nose is impacted by several de-mographic characteristics ofinfants and can be used to predict microbiomecomposition. Furtherstudies are needed to design appropriate algorithms that are able to predict NEC based on fecalVOC profiles. ◎ 2020 Elsevier Inc. All rights reserved. ( This article was presented at the Academic Su r gical Congress， February 2020，Orla n do， FL. ) Introduction Volatile organic compounds (VOCs) have recently emerged asa potential noninvasive biomarker in many diseases. VOCs aremetabolites that produce a characteristic odor and are pro-duced by both normal and pathophysiological processes.Clinicians have historically used their own noses to assessVOCs to aid in the diagnosis of diabetic ketoacidosis， charac-terized by an acetone"fruity"odor in a patient’s breath. It didnot take long for researchers to use the strong sense of smellof canines to diagnose many diseases， including severaldifferent types of cancers and Clostridium difficile colitis.-6 Theanalysis of VOCs has become more sophisticated and is nowdone by several analytical methods， such as electronic nose(eNose) software， gas chromatography， mass spectrometry，and headspace solid-phase microextraction. This advancement in the analysis of VOCs has led to severalstudies in researching their ability to predict disease. Stoolanalysis of VOCs has been performed to identify gastrointes-tinal diseases. Fecal VOCs are heterogenous compounds thatoriginate from metabolic byproducts from bacteria. They havebeen used to diagnose infectious processes such as Clostridiumdifficile colitis and Campylobacter jejuni enteritis. The hostintestinal tract also produces VOCs and has been used to di-agnose noninfectious diseases， such as neoplastic and inflam-matory processes. There have been several case-controlstudies using VOCs as a noninvasive biomarker for earlydetection of colorectal cancer and advanced adenomas.9，10However，there was significant variability of technologies usedto assess VOCs in these studies，and further study is needed toidentify specific biomarkers to colorectal cancer that are drivingchanges in VOCs.In addition， fecal VOCs have been used fordiagnosis of inflammatory bowel disease and irritable bowelsyndrome and its response to dietary modification.13，14 Given the advancement of VOCs in the diagnosis ofgastrointestinal diseases， it is important to consider whetherVOC analysis by an eNose device may be used in the diagnosisof necrotizing enterocolitis (NEC)， which remains a seriousand devastating disease that affects the gastrointestinal tractof the newborn.15 NEC often presents clinically as abdominaldistention， feeding intolerance， and eventual rapid progres-sion to sepsis and surgical intervention requiring resection ofthe necrotic intestine.16，17 Despite decades of research， theincidence of NEC remains high at 5%-7% of all preterm neo-nates.18 In addition， mortality remains unabated and is highas 42% in very low birth weight infants.1 Although there have been two previous studies that havedemonstrated that the eNose device can be used to diagnoseNEC， they did not take into account varying patient de-mographics that could have an impact on VOCs.20，21 Beforethe development of a unique“NEC” VOC signal， it is importantto assess whether eNose device analysis will be skewed bycertain demographics. We hypothesized that fecal VOC sig·nals produced by the Cyranose 320 eNose device (Sensigent，Baldwin Park， CA； Fig. 1) would vary based on hospital loca-tion， sex， diet， and caloric content in healthy infants. Inaddition， we hypothesized that VOC profiles would correlateto the intestinal microbiome and could be used via machinelearning to predict microbial enterotype. Materials and methods Patient enrollment This prospective study was performed from July 2018 toJanuary 2019. This study was approved by our local institu-tional review board with waiver of informed consent (IRB Fig. 1 -Cyranose 320 eNose device was used to analyze fecal VOCs from infant stool.(Color version of figure is available online.) Hospital 1 (7) Hospital 2 (8) Hospital 3 (5) Hospital 4 (11) Sex Male， n (%) 4(57) 3(38) 3(60) 5(45) 0.832 Gestational age， median 39+0(36+4 to 39+0(38+1 to 37+1(36+2to 42+0 (38+ 3 to 0.033 (IQR)， wk +d 41+4) 39+5) 37+5) 45+0) Feeding pattern，n (%) Breast milk ± formula 5(71) 7(88) 2(40) 3(27) 0.051 Exclusive formula 2(29) 1(12) 3 (60) 8(73) 0.051 IQR=interquartile range. protocol number 1803568191). Fresh stool was collected frominfants in the newborn intensive care unit (NICU) at fourhospitals. Samples from three of the four hospitals wereanalyzed within 1 h of collection. One hospital was located asignificant distance away from the site of analysis， and sam-ples from that site were placed on ice before being analyzed.Patients eligible for enrollment were term， healthy infants onenteral feeds. Infants who had advanced medical comorbid-ities or who were on antibiotics were excluded from the study.Demographic data recorded included gestational age，sex， diet(formula， breastmilk， and both)， caloric content， and hospitallocation. The caloric density of breastmilk was assumed to be20 kcal/oz， and caloric density of formula and formula sup-plements were known and recorded. VOC analysis by eNose Fresh stool samples were analyzed using the Cyranose 320eNose device， as previously described.Approximately 1 g ofstool was placed in a sealed container and slowly heated to37°C to enhance VOC signal from the samples and mimic theinternal temperature of infants. Researchers were blinded toclinical data when performing analysis. The container wasconnected to the eNose device using standard oxygen tubing，and stopcocks were used to create a closed-loop circuit toprevent ambient air from interfering with analysis. Baselinereference signal was created by connecting the eNose deviceto an organic vapor air filter (Honeywell Part No. 7581P100L).Samples were then analyzed in random order using theCyranose 320 eNose device.The Cyranose 320 uses a NoseChiparray， which contains up to 32 polymer nanocomposite car-bon sensors. These individual sensors react with VOCs andundergo a change in resistance that is uniquely related to theVOCs biochemical profile， creating a unique “smellprint."Using advanced pattern recognition， the eNose can be taughtto discriminate between samples. 16s rRNA gene sequencing Genomic DNA was extracted， and the V4 region of the bacte-rial 16S rRNA gene was amplified using the Shoreline Com-plete V4kit (Shoreline Biome). The DNA library was sequencedon an iSeq 100 with 150-bp paired-end reads. Sequences werecurated using the mothur software package. Paired-end reads were merged into contigs， screened for sequencing errors， andaligned to the SILVA bacterial SSU reference database. Alignedsequences were screened for chimeras and classified usingthe Greengenes database. Samples were binned into com-munity types using k-means clustering applied to the relativeabundances of bacterial families. Statistical analyses All statistical analyses were performed using R software (Rversion 3.5.1， R Foundation for Statistical Computing， ViennaAustria). Ordinations of smellprints were constructed usingprincipal component analysis. Overall smellprints wereassociated with clinical variables usinga PERMANOVA test， asimplemented in the adonis function from the vegan R pack-age. Individual sensors were associated with clinical variablesusing a Mann-Whitney U-testforbinaryvvariables.Kruskal-Wallis test for categorical variables， and Spearmancorrelation for continuous variables. Gut community typeswere associated with individual Cyranose sensors usingKruskal-Wallis tests. The three sensors significantly associ-ated with gut community types were used to construct fast-and-frugal tree models using the FFTrees package in R. Results Demographics A total of 31 infants were enrolled in the study. Demographicsof patients are shown in Table 1. Sex distribution of infantsenrolled was equivalent across all sites (P=0.832). Infantsenrolled at Hospital 4 were older than infants enrolled at othersites (P =0.033). The proportion of infants who receivedbreastmilk and those who consumed exclusive formula wasroughly equivalent across all groups (P =0.051). Diet and caloric intake The overall smellprints were significantly different for infantsconsuming human breastmilk compared with those who wereconsuming exclusively formula (P=0.032； Fig.2A). In addition，nine of 28 individual sensors were significantly different forinfants who consumed breastmilk compared with those who Fig. 2-Differences in fecal VOCs between diet， calories， sex， and hospital location. (A) Overall smellprints were significantlydifferent between infants consuming exclusive human breastmilk (HBM) and infants consuming exclusive formula. (B)Sensor 5 and 25 were the most different. (C) Overall smellprints were not significantly different according to caloric intake.(D) Sensor 16 had the strongest negative correlation， and sensor 12 had the strongest positive correlation between calories.(E) Overall smellprints were not significantly different between male and female infants. (F) Two sensors were significantlydifferent. (G) Overall smellprints were not significantly different between four hospital locations. (H) Sensors 11 and 23 werethe most different based on hospital of birth. (Color version of fgure is available online.) MDS Axis 1 Fig.3 - 16s rRNA analysis of stool samples. (A) Relative abundance of bacterial families grouped by enterotype. (B)Nonmetric multidimensional scaling ordination based on Bray-Curtis dissimilarity of family abundances. The microbiotacomposition was significantly different between enterotypes (P=0.00002). (Color version of figure is available online.) consumed formula (Fig. 2B). When examining the caloricdensity of the feeds， overall smellprints did not significantlydiffer (P=0.22； Fig.2C). Two sensors individually correlatedwith calories (Fig. 2D). Sex and hospital of birth Overall smellprint analysis by Cyranose 320 eNose device didnot show a significant difference between male and femaleinfants (P=0.3； Fig. 2E)； however， two individual sensorsdiffered according to infant sex (Fig. 2F). In addition， overallsmellprints did not significantly differ for infants based onhospital location of birth (P=0.52； Fig.2G). Although a longeramount of time passed between specimen collection andanalysis for Hospital 1， there was still no significant differencein smellprint between hospital location. Three individualsensors， however， were significantly different based on thehospital of birth (Fig. 2H). Microbiome analysis Microbiome analysis by 16s rRNA sequencing demonstratedthat each infant's microbiome was dominated by one of threedistinct families： Bifidobacteriaceae， Enterobacteriacae， andEnterococcaceae. These three families accounted for >75% of allsequences and were observed across different hospital loca-tions (Fig. 3A). These observations are consistent with theexistence of three infant microbiome enterotypes， whichdiffer greatly from the enterotypes proposed in adults both interms of the defining bacterial taxa (Bacteroides， Prevotella， andRuminoccus in adults) and the magnitude of difference betweeneenterotypes.23-25 Nonmetric multidimensionalscaling ordination was used to visualize the relationship be-tween microbiomes based on Bray-Curtis dissimilarity offamily abundances (Fig. 3B). Samples in the ordination clus-tered into three distinct groups， consistent with the proposedenterotypes. Despite the extreme differences in microbiomecomposition between enterotypes， they were not significantlyassociated with any of the infant demographics (Table 2). Analysis by the Cyranose 320 eNose device grouped byenterotype showed that three smellprint sensors weresignificantly associated with certain enterotypes (Fig.4). Tofurther examine the ability of smellprint sensors to differen-tiate between enterotype， a decision tree model was con-structed using the fast-and-frugal trees algorithm (Fig.5). Thesimple model could differentiate the two most commonenterotypes， Enterotype I and Enterotype 2， with 88% accuracyusing only two sensors， suggesting an eNose could be used torapidly determine the enterotype of an infant’s microbiome. Table 2 -Enterotype association with patient demographics. Metadata P value Diet 0.21 CaloriesT 0.58 Sex 0.63 Hospital 0.48 Gestational age 0.97 Chi-squared test. TAnalysis of variance. 6oc Fig. 4-Three individual sensors were significantly associated with enterotype. (Color version of figure is available online.) Fig. 5-Decision tree model created by the fast-and-frugal tree algorithm was used to test whether smellprint sensors couldbe used to predict the two most common enterotypes.The top panel shows the training set population， which consisted of12 samples with microbiomes belonging to Enterotype 1 (squares) and 14 from Enterotype 2 (triangles). The middle panelshows the structure of the decision tree， which splits samples based on the readings from two smell print sensors. Greenshapes indicate correct classifications， whereas red shapes indicate erroneous classifications. The bottom panelsummarizes the performance of the model using a variety of metrics，including the error matrix， sensitivity， specificity，accuracy， and a receiver operator characteristic (ROC) curve. (Color version of figure is available online.) Discussion NEC is a devastating abdominal disease in infants， and to date，there are no adequate biomarkers to aid clinicians in diag-nosis. In this study， we found that eNose technology could beused to discriminate infants based on certain demographicfactors. Overall smellprints differed for infants consumingbreastmilk compared with those exclusively consuming for-mula. In addition， we found that several sensors were affectedby variations in calories， the hospital where the infant wasborn， and by the infant’s sex. There have been two proof of principle studies alreadycompleted on whether fecal VOCs can be used to detectNEC. The first study by Garner et al. was a single centerstudy that used frozen stool samples from NICU infants.Stool was heated， and VOCs were analyzed by solid-phasemicroextraction， gas chromatography， and mass spectrom-etry. This retrospective study demonstrated that a change inVOC profile predated diagnosis of NEC by up to 4 d.2 Thesecond study was a multicenter study performed in theNetherlands on frozen stool using the same Cyranose 320eNose device technology. This study also demonstrated achange in VOC profile predating the diagnosis of NEC whencompared with healthy controls.21 Our study contributes tothese findings by highlighting the importance of consideringinfant demographics when performing fecal VOC analysis.Given that variables such as hospital location and diet canaffect eNose sensors， the characteristic“NEC VOC output”may be quite different depending on the geographic locationof the patient. In addition， this same output may be influ-enced by whether or not the infant is on formula feeds orbreastmilk. Adjusting for these variables will be crucial inaccurately detecting NEC by VOC analysis and may involveadjusting the sensitivity of sensors that are affected by thesevariables or creating an algorithm that creates an “NECscore”that appropriately weighs the impact of these con-founding variables. Our study found that using a simple decision tree model，the Cyranose 320 eNose device could reliably differentiatebetween common stool bacterial communities. These com-munity types or “enterotypes"represent a distinct stoolcommunity type based on the relative abundance of bacterialfamily.23-25DDistinct enterotypes have been shown to beassociated with different pathologies， including obesity andinflammatory bowel disease.26 In addition， as our under-standing of NEC evolves， it is becoming clear that NEC ispreceded by intestinal dysbiosis， and there are certain pre-dictable patterns of microbial change that increase the risk ofNEC.27-30 Given that the Cyranose 320 eNose device can pre-dict certain enterotypes， it may also be able to identify subtlechanges in the microbiome that predispose an infant todeveloping NEC. This potential for early detection of intestinaldysbiosis by VOC analysis is particularly exciting. As thisassay is fast and relatively inexpensive， it would allow forearlier cessation of enteral feeds and initiation of antibiotics.Identifying these microbial changes by 16s rRNA sequencingand shotgun metagenomics is often costly and timeconsuming， which is unfortunately unforgiving in a rapidlyevolving and often fatal disease. A strength of this study is the prospective design thatincluded infants from multiple different centers around theIndianapolis area. In addition， this study identified majorbacterial families in the microbiome of healthy infants by 16srRNA sequencing and correlated it to fecal VOC analysis.Finally， the stool was analyzed fresh and in real time by theCyranose 320 eNose device， which closely mimics the pro-posed translational model where fecal VOCs from a diaper inthe NICU could be analyzed quickly by nursing staff to detectNEC. There are limitations to this study. First， although infantswere enrolled from four different centers， there were only 31patients enrolled in the study. Second， the samples were onlyanalyzed in real time and were not analyzed after beingfrozen. It has been demonstrated that storing fecal samples indifferent settings can increase the variability of fecal VOCanalysis，3and in order to make this study more generalizableand reproducible， it would be helpful to compare fresh versusfrozen analysis by the eNose device. Third， for the decisiontree model shown in Figure 5， there were not enough samplesto develop both a validation and training set. Collection ofmore samples will be necessary to validate the tree modelshown in the future.Finally，although we are able to establishthat different conditions such as consuming breastmilkchange the overall smellprint of the eNose device， we areunable to establish which precise metabolites caused thischange. Although the eNose device could theoretically iden-tify a known biomarker in a pure sample， stool from an infantwith NEC is heterogenous and has many VOCs， with metab-olites both from the host and bacteria， all of which contributeto the smellprint. Concurrent analysis with gas chromatog-raphy or stool proteomics to identify and quantify metaboliteswould be helpful， as it would provide additional insight intothe underlying mechanism of why breastmilk and differentfamilies of bacteria produce different signals on the polymernanocomposite carbon sensors. Conclusions In summary， fecal VOC analysis by the Cyranose 320 eNosedevice can be used to predict microbial enterotype and isaffected by patient hospital location，sex， and diet.Fecal VOCanalysis by this device is rapid and inexpensive and may beperformed at the bedside as a biomarker for the risk ofdeveloping NEC. Further studies are needed to design appro-priate algorithms that are able to predict NEC based on fecalVOC profiles. Acknowledgment T.A.M. was supported by K08DK113226 from the National In-stitutes of Health， the George H. Clowes Memorial ResearchCareer Development Award， the Koret Foundation， and theDepartment of Surgery at the Indiana University School ofMedicine. Authors’contributions： B.D.H. and A.R.P. performed stoolacquisition and data interpretation and drafted the article. N.T.B. and T.B.H. performed Cyranose analysis， statisticalanalysis， and data interpretation. T.A.M. provided guidance，project insight，and data interpretation. All authors providedcritical revisions to the manuscript and approved of its finalform of submission. Disclosure The authors reported no proprietary or commercial interest inany product mentioned or concept discussed in this article. REFERENCES 1. Ahmed I， Niaz Z， Ewbank F， Akarca D， Felwick R， Furnari M.Sniffing out causes of gastrointestinal disorders： a review ofvolatile metabolomic biomarkers. Biomark Med.2018；12：1139-1148. 2. Kumar V， Abbas AK， Aster JC. Robbins and Cotran pathologicbasis of disease. Philadelphia，PA： Elsevier/Saunders； 2015：1391. 3. McCulloch M， Jezierski T， Broffman M， Hubbard A， Turner K，Janecki T. Diagnostic accuracy of canine scent detection inearly- and late-stage lung and breast cancers. Integr CancerTher. 2006；5：30-39. 4. Sonoda H， Kohnoe S， Yamazato T， et al. Colorectal cancerscreening with odour material by canine scent detection. Gut.2011；60：814-819. 5. W\illis CM， Church SM， Guest CM， et al. Olfactory detection ofhuman bladder cancer by dogs： proof of principle study. BMJ.2004；329：712. 6. BEomers M， Van Agtmael M， Luik H， Van Veen M，Vandenbrouchke-Grauls C， Smulders Y. Using a dog'ssuperior olfactory sensitivity to identify Clostridium difficilein stools and patients： proof of principle study. BMJ.2012；345：e7396. 7. Dixon E， Clubb C， Pittman S， et al. Solid-phasemicroextraction and the human fecal VOC metabolome. PLoS0ne. 2011；6：e18471. 8. Garner CE， Smith S， de Lacy Costello B， et al. Volatile organiccompounds from feces and their potential for diagnosis ofgastrointestinal disease. FASEB J. 2007；21：1675-1688. 9. ABosch S， Berkhout DJ， Ben Larbi I， de Meji TG， de Boer NK.Fecal volatile organic compounds for early detection ofcolorectal cancer： where are we now? J Cancer Res Clin Oncol.2019；145：223-234. 10. de Meij TG，Larbi IB， van der Schee MP，et al. Electronic nosecan discriminate colorectal carcinoma and advancedadenomas by fecal volatile biomarker analysis： proof ofprinciple study. Int J Cancer. 2014；134：1132-1138. 11. Di Lena M， Porcelli F， Altomare DF.Volatile organiccompounds as new biomarkers for colorectal cancer： areview. Colorectal Dis. 2016；18：654-663. 12. de Boer NK， de Meji TG， Oort FA， et al. The scent of colorectalcancer： detection by volatile organic compound analysis. ClinGastroenterol Hepatol. 2014；12：1085-1089. 13. de Meij TG， de Boer NK， Benninga MA， et al. Faecal gasanalysis by electronic nose as novel， non-invasive method forassessment of active and quiescent paediatric inflammatorybowel disease： proof of principle study. J Crohns Colitis. 2014. 14. Rossi M， Aggio R， Staudacher HM， et al. Volatile organiccompounds in feces associate with response to dietaryintervention in patients with irritable bowel syndrome. ClinGastroenterol Hepatol. 2018；16：385-391 e1. 15. Lin PW， Stoll BJ. Necrotising enterocolitis. Lancet.2006；368：1271-1283. 16. Kliegman R. Nelson textbook of pediatrics. 21st ed 2.Philadelphia， PA： Elsevier； 2020：I1-I140. 17. Ashcraft K， Holcomb G， Murphy Patrick. Ashcraft's PediatricSurgery. 5th ed. Philadelphia， PA： Saunders/Elsevier； 2010. 18.Z4ani A， Pierro A. Necrotizing enterocolitis： controversies andchallenges. F1000Res. 2015：4. 19. Fitzgibbons SC， Ching Y， Yu D， et al. Mortality of necrotizingenterocolitis expressed by birth weight categories.J PediatrSurg. 2009；44：1072-1075 [discussion 1075-6]. 20. Garner CE， Ewer AK， Elasouad K， et al. Analysis of faecalvolatile organic compounds in preterm infants who developnecrotising enterocolitis： a pilot study. J Pediatr GastroenterolNutr. 2009；49：559-565. 21. de Meij TG， van der Schee MP， Berkhout DJ， et al.Earlydetection of necrotizing enterocolitis by fecal volatile organiccompounds analysis. J Pediatr. 2015；167：562-567 e1. 22. Rock F， Barsan N， Weimar U. Electronic nose： current statusand future trends. Chem Rev. 2008；108：705-725. 23. Koren O， Knights D， Gonzalez A， et al. A guide toenterotypes across the human body： meta-analysis ofmicrobial community structures in human microbiomedatasets. PLoS Comput Biol. 2013；9：e1002863. 24.DIing T， Schloss PD. Dynamics and associations of microbialcommunity types across the human body. Nature.2014：509：357-360. 25. Arumugam M， Raes J， Pelletier E， et al. Enterotypes of thehuman gut microbiome. Nature. 2011；473：174-180. 26.(Quince C， Lundin EE， Andreasson AN， et al. The impact ofCrohn’s disease genes on healthy human gut microbiota： apilot study. Gut. 2013；62：952-954. 27.1Warner BB， Deych E， Zhou Y， et al. Gut bacteria dysbiosisand necrotising enterocolitis in very low birthweightinfants： a prospective case-control study. Lancet.2016；387：1928-1936. 28. Mai V， YoungCM， Ukhanova M， et al. Fecal microbiota inpremature infants prior to necrotizing enterocolitis. PLoS One.2011；6：e20647. 29. Zhou Y， Shan G， Sodergren E， Weinstock G， Walker WA，Gregory KE. Longitudinal analysis of the premature infantintestinal microbiome prior to necrotizing enterocolitis： acase-control study. PLoS One.2015；10：e0118632. 30. Sim K， Shaw AG， Randell P， et al. Dysbiosis anticipatingnecrotizing enterocolitis in very premature infants. Clin InfectDis. 2015；60：389-397. Bosch S， El Manouni El Hassani S， Covington JA， et al.Optimized sampling conditions for fecal volatile organiccompound analysis by means of field asymmetric ionmobility spectrometry. Anal Chem. 2018；90：7972-7981. 粪便挥发性有机化合物（VOC）的评估已在许多不同的病理学中作为一种非侵入性生物标记物出现。在评估VOC是否可用于诊断包括坏死性小肠结肠炎（NEC）在内的肠道疾病之前，有必要测量可变的婴儿人口统计学因素对VOC信号的影响。从大都市区的四家医院的足月婴儿中收集粪便样本。加热样品，并通过Cyranose 320电子鼻评估粪便中的挥发性有机化合物。28个传感器组合成一个整体的气味印记，并且也进行了单独评估。使用16s rRNA基因测序对婴儿微生物组进行分类。气味印记与进食类型（配方奶与母乳），性别，出生医院和微生物肠型相关。通过PERMANOVA用欧几里德距离评估总体气味印记，并通过Mann-Whitney U检验评估每个气味印记的单个传感器。P  <0.05显着。根据饮食习惯，总体气味印记差异显着。各个传感器根据性别和出生医院的不同而有显着差异，但总体气味印记没有明显差异。使用矩树模型，两个单独的传感器可以可靠地预测微生物肠型。用电子鼻评估粪便中的挥发性有机化合物受到婴儿的几种人口统计学特征的影响，可用于预测微生物组的组成。需要进一步的研究来设计能够基于粪便VOC曲线预测NEC的合适算法。