Early prediction of bronchopulmonary dysplasia by urinary metabolomics: a case–control study

Abstract

Objective

Bronchopulmonary dysplasia (BPD), the most frequent complication of extreme preterm birth, lacks not only of a comprehensive definition but also of effective treatments and predictive tools. Metabolomics is a valuable tool to unravel the underlying pathogenetic pathways of diseases and identify possible early markers. The objective of this study was to find metabolic signatures of subsequent BPD development, defined and stratified as per Jobe and Bancalari 2001 NHICD Consensus.

Methods

In this observational case–control study, we initially enrolled 161 very preterm unmatched infants, collected their urine samples during the first 24 hours of life and performed metabolomics evaluations on these samples. Patients were then followed until 36 weeks postmenstrual age. To reduce the influence of gestational age and other confounders on metabolome, we applied a nested case–control matching procedure that allowed the selection of 25 BPD cases and 25 non-BPD controls.

Results

Multivariate and univariate data analysis led to the recognition of 17 metabolites related to BPD development in the first day of life, of which three were identified: L-Glutamic acid (p value=0.038), o-Hydroxyphenylacetic acid (p=0.039), L-Homoserine (p value=0.020). Some of these metabolites are known to play a role in the protection against oxidative stress and/or inflammatory response, two of the most known factors involved in BPD pathogenesis. In particular, L-Glutamic acid and its ionic form glutamate were increased in infants developing BPD suggesting a role as promising marker of the disease.

Conclusions

Our findings pave the way to better characterise early origin of BPD from a metabolic point of view towards a better biological framework of the disease and, eventually, its prediction and possible new treatments.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• The prediction of bronchopulmonary dysplasia (BPD) is not easy but would be essential for its prevention. Metabolomics shows a promising potential in this field, but with only small studies exploring it.

WHAT THIS STUDY ADDS

• In a larger case–control study, the metabolic fingerprint of BPD development is already evident in the first day of life. Glutamate in particular is the most promising marker.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• Early prediction of BPD could allow an early implementation of new preventive drugs currently under evaluation. Moreover, the identified markers could suggest new pathogenetic pathways and therapeutic fields to be explored.

Background

Bronchopulmonary dysplasia (BPD) is a long-lasting respiratory complication of preterm birth,¹ with known comorbidities and possible consequent morbidity, health-related costs and early mortality.^{2 3} The incidence of BPD is actually stable or even increasing⁴ due to the improved neonatal care that allowed an increased survival of extreme preterm infants, combined with the absence of effective and safe preventive or therapeutic options.

The early prediction of BPD would be crucial to implement known and new therapeutic options to prevent the development of this disease;⁵ moreover, the identification of specific pathways involved in BPD development would allow the development of targeted preventive therapies.⁶

The actual options to predict BPD are principally based on clinical data like gestational age at birth, and birth weight,⁷ but these are not directly related to the pathogenesis of the disease. Consequently, similar preterm infants can be identified as having similar BPD risk, even if the disease develops in one child and not the other.

Metabolomics, a high-throughput technique, is gaining importance as a possible tool to unravel the underlying pathogenetic pathways of diseases and consequently identify possible early biomarkers in the prediction of the disease. Metabolomics was chosen for this study because it offers a direct and dynamic snapshot of physiological and pathological processes, especially relevant when investigating complex, multifactorial diseases such as BPD. Specifically, metabolomics provides a real-time readout of actual cellular activity, capturing the downstream effects of gene expression and protein function. Metabolomics is also particularly suited for early biomarker discovery, as metabolites often change before phenotypic symptoms become clinically apparent.

Metabolomic application in BPD begins in 2014 with Fanos et al⁸ that showed a different urinary metabolic profile of the first 2 days of life in those preterm infants (<29 weeks of gestation) that developed BPD compared with controls. In particular, they interpreted these findings as signs of anaerobic metabolism and activation of protective pathways. The same group⁹ performed urinary metabolomics also at 7 days of life of preterm infants (<32 weeks of gestation) identifying some discriminating metabolites possibly involved in the pathogenetic pathways of BPD.

The biomarkers of BPD development can also be found in the amniotic fluid as shown by our group¹⁰ that identified a distinctive metabolic profile in that of those infants who subsequently were born preterm and developed BPD.

Piersigilli and colleagues¹¹ chose to analyse the metabolome in tracheal aspirates of preterm infants of less than 30 weeks of gestation identifying 53 metabolites characteristic of BPD and possibly involved in its pathogenesis, including increased levels of some amino acids, taurine, spermidine and some acylcarnitines.

Other sample sources were also explored like umbilical cord blood,¹² dried blood spots,¹³ stool¹⁴ and exhaled breath condensate¹⁵ with encouraging results confirming the possibility of early intercepting the development of BPD and the fact that multiple pathways are involved in its pathogenesis (eg, amino acids, lipids, carbohydrates, organic acids, acyl-carnitines and many others).

This evidence has been recently summarised by Guo et al¹⁶ in their systematic review and meta-analysis, in which they recognised how amino acids are the key metabolites distinguishing preterm infants with and without BPD, with glutamate (showed as increased in tracheal aspirates and dried blood spots) potentially serving as a BPD predictor.

The aim of this study was to apply the untargeted metabolomic approach to urine samples, collected in the first 24 hours of life, to discover the presence of an early metabolic signature of BPD development at birth in preterm infants of the same gestational age and secondarily to identify those early metabolic pathways involved in the pathogenesis of BPD to better characterise the disease and drive the development of potential therapeutic treatments.

Methods

Population

All preterm infants born at less than 32 weeks of gestation admitted at birth to the level IV NICU of the University Hospital of Padova, Italy between March 2015 and June 2020 were screened for enrolment in this monocentric observational case–control study. Those with major congenital or chromosomal abnormalities or with a suspected metabolic disease were excluded. The remaining were enrolled after obtaining parental informed consent.

The infants were assigned to cases (BPD) or controls group (non-BPD) following the definition criteria by Jobe and Bancalari¹⁷ from 2001 NICHD consensus. Cases were stratified for severity after reaching 36 weeks PMA or discharge following the same definition.¹⁷

For each enrolled subject clinical data were collected from medical records and seven neonatal parameters related to BPD development⁵ were recorded for case–control matching: sex, gestational age at birth, birth weight, intrauterine growth restriction, premature rupture of membranes, early onset of sepsis and antenatal steroid treatment.

Sample collection and preparation

For each subject, a non-invasive collection of at least 2 mL of urine was performed by inserting a sterile cotton swab inside the newborn’s diaper and checking for the presence of urine every 30 min until the first urination, always within 24 hours of life, as previously described.¹⁸ Samples were subsequently placed in the refrigerator and then frozen as soon as possible, within 2 hours to avoid any metabolite degradation or transformation of metabolites.¹⁹

Metabolomic analysis

Metabolomics analysis was performed at the Mass Spectrometry and Metabolomics Laboratory of the Women’s and Children’s Health Department, University of Padova, Padova, Italy. Untargeted metabolic profiling was performed in positive and in negative electrospray ionisation mode on an Acquity Ultra Performance Liquid Chromatography (UPLC) system (Waters, UK) coupled to a Quadrupole Time-of-Flight Synapt XS HDMS mass spectrometer (Waters MS Technologies, Manchester, UK). The mass range scan was of 20–1200 amu, in MS scan mode. For LC-MS analysis, a Waters Acquity UPLC HSS T3 column 2.1 mm wide and 100 mm long packed with 1.8 µm beads was used and its temperature was kept at 50°C. The chromatographic method was the same reported in Priante et al.²⁰

Data preprocessing

Raw data were extracted using Progenesis QI software V.3.0 (Waters Corporation, USA). The parameters were optimised through the preliminary processing of the quality controls samples (QCs). Specifically, 0.5 was set as filter to import the raw data, and the QC in the middle of the sequence was selected as reference for the automatic retention time alignment of the samples in the sequence. The sensitivity of the automatic algorithm for peak picking was set at 5 in the time range from 0.4 min to 8.0 min. As a result, the so-called time@mass variables (where ‘time’ is the retention time and ‘mass’ is the mass to charge ratio of the spectral feature) were generated. Features with at least one missing data in the QCs or more than 10% of missing data in the samples were eliminated. For each variable passing such a filter, missing data were imputed with a random number between zero and the minimum value measured for that variable. Data were calibrated on the basis of the local linear regression models obtained considering the trend of the QCs at different dilution with the run order. Probabilistic quotient normalisation was applied to take into account dilution effects. Variables with a coefficient of variation greater than 30% in the QCs were excluded.

Statistical data analysis

The clinical data of the recruited neonates were investigated by t-test for normally distributed data, Mann-Whitney test for non-normally distributed data, and Yates’s χ² test for categorical data, assuming a significance level α=0.05. Normality was assessed using the Shapiro-Wilk test assuming normally distributed data for p value greater than 0.10.

Both multivariate and univariate data analysis were applied to investigate metabolic data. Specifically, exploratory data analysis and outlier detection were performed by principal component analysis (PCA), whereas partial least square (PLS) for classification with a two-component model (PLS2C)²¹ was employed to solve the classification problem, where BPD and non-BPD subjects were compared. PLS model was post-transformed to simplify model interpretation,²² whereas stability selection based on variable influence on projection score (500 random submodels were generated and a significant level α=0.05 was assumed) was applied to discover relevant features and to estimate the performance in prediction of the models.²³ Permutation test was performed to assess model reliability (1000 random permutations were generated). The number of PLS-score components to use was estimated on the basis of the maximum Matthews correlation coefficient (MCC) calculated by 20-repeated 5-fold cross-validation (MCC_CV) under the condition to pass the permutation test. Logistic regression was applied to investigate how each single metabolic feature can predict the BPD outcome. Logistic models were corrected for gestational age.²⁰ Data were autoscaled prior to performing data analysis.

Since gestational age and other clinical data may strongly affect the urinary metabolome, to reduce biases in data modelling, two groups of patients, one composed of neonates developing BPD and one of controls (non-BPD), matched for the recorded clinical data (sex, gestational age at birth, birth weight, intrauterine growth restriction, premature rupture of membranes, early onset of sepsis and antenatal steroid treatment) were extracted applying the same procedure precisely described in Priante et al²⁰ and summarised hereafter. The distance matrix between the eligible neonates of the BPD and of the non-BPD groups was calculated considering the recorded clinical data. For each BPD neonate, the non-BPD neonate with the minimum distance was selected, and the pairs of case–control obtained were sorted according to the increasing distance. The pairs with the greatest distance were iteratively excluded until the set of the selected cases and that of the selected controls showed p values greater than a predefined threshold in their clinical data. More details can be found in Section S2 of Supplementary Materials of Sumner et al.²⁴

Data analysis was performed by in-house R-functions implemented using the R V.4.0.4 platform (R Foundation for Statistical Computing, Vienna, Austria).

Variable annotation

The relevant variables selected by data analysis were annotated on the basis of the mass to charge ratio (m/z) and the retention time (Rt) by searching our in-house database of commercial standards, the METLIN metabolite database and the Human Metabolome Database using 4 different levels of confidence according to Sumner et al.²⁴ Specifically, annotation level 1 (identified metabolite) was assigned to compounds with a difference of less than 10 ppm for m/z and 0.2 min for Rt with respect to standards of our in-house database, which were analysed under identical conditions to the current analysis, level 2 was attributed to metabolites with m/z less than 10 ppm and with similar fragmentation patterns with respect to the online databases, level 3 to compounds with m/z less than 10 ppm on the online databases, and level 4 for repeatable signals of mass spectrum, with no annotation in the databases used.

Results

Initially, urine samples were collected in an unmatched population of 161 neonates during the first 24 hours of life, 69 of them were diagnosed with BPD (42.9%) and 92 without BPD (57.1%) at 36 weeks PMA. Among neonates with BPD, 36 (52.2%) were classified as mild BPD, 16 (23.2%) as moderate BPD, 17 (24.6%) as severe BPD. The clinical characteristics of the neonates are reported in table 1.

Table 1. Clinical data of the recruited neonates; numerical data non-normally distributed are reported as median (IQR) and categorical data as absolute number (prevalence).

Clinical data	BPD group (n=69)	non-BPD group (n=92)	P value
Male sex	41 (59.4%)	45 (48.9%)	0.24*
Gestational age at birth (weeks)	26.14 (2.71)	30.14 (2.00)	<0.001†
Birth weight (g)	760 (330)	1335 (340)	<0.001†
Intrauterine growth restriction	12 (17.4%)	7 (7.6%)	0.09*
Premature rupture of membranes>18 hour	16 (23.2%)	37 (40.2%)	0.04*
Early onset sepsis	27 (39.1%)	21 (22.8%)	0.04*
Antenatal steroid treatment	63 (91.3%)	88 (95.7%)	0.42*
Duration of mechanical ventilation (days)	8 (2–26)	0 [0–1)	<0.001*

^*Yates’ χ2 test.

^†Mann-Whitney test.

BPD, bronchopulmonary dysplasia.

The two groups were significantly different with respect to gestational age, birth weight, premature rupture of membranes, duration of mechanical ventilation and early onset of sepsis. Specifically, neonates developing BPD showed both lower gestational age and lower birth weight than controls (table 1).

To better investigate if a urinary metabolic signature may early predict BPD, the effects of the gestational age and birth weight were controlled retrospectively matching the cases and controls. Applying the algorithm for matching described in Priante et al²⁰ and assuming a threshold of 0.05 for the p values, a subset of 25 neonates developing BPD and a subset of 25 matched controls were selected. In table 2, the clinical data of the two subsets are reported. The only significantly different aspect between these two populations remained, as expected, the duration of mechanical ventilation.

Table 2. Clinical data of the selected matched neonates; numerical data non-normally distributed are reported as median (IQR) and categorical data as absolute number (prevalence).

Clinical data	BPD group (n=25)	non-BPD group (n=25)	Pvalue
Male sex	13 (52.0%)	15 (60.0%)	0.78*
Gestational age at birth (weeks)	27.71 (2.14)	28.28 (3)	0.07†
Birth weight (g)	1000 (240)	1120 (310)	0.12†
Intrauterine growth restriction	4 (16.0%)	4 (16.0%)	1.00*
Premature rupture of membranes>18 hour	6 (24.0%)	6 (24.0%)	1.00*
Early onset of sepsis	6 (24.0%)	5 (20.0%)	1.00*
Antenatal steroid treatment	24 (96.0%)	24 (96.0%)	1.00*
Duration of mechanical ventilation (days)	3 (1;7)	0 (0;1)	<0.001†

^*Yates’s χ2 test.

^†Mann-Whitney test.

BPD, bronchopulmonary dysplasia.

After data preprocessing, a dataset composed of 740 metabolic features was obtained. No outliers within each group were detected by PCA on the basis of the T2 and Q-tests assuming a significance level of 0.05.

Applying stability selection, 10 variables were selected as relevant. The PLS model obtained considering the selected variables showed three score components, MCC=0.800 (p value=0.005) and MCC_CV=0.560 (p value=0.005). The predictive component did not result to be correlated to gestational age and birth weight, or to other recorded clinical data. Moreover, the median of the MCC for the out-of-bag prediction was 0.570 (corresponding to a median AUC of the ROC curve equal to 0.88), showing that the model is promising in predicting BPD.

Univariate data analysis based on logistic regression discovered 15 metabolic features with p value less than 0.05. Merging the results of multivariate and univariate data analysis, a set of 17 relevant metabolic features were obtained, 8 out of which were annotated at level 1, 2 or 3 (table 3). The boxplots representing the distributions of the three identified metabolites, that is, features with annotation level 1, within the two groups are reported in figure 1.

Table 3. Annotation of the 17 relevant metabolites discovered merging the results of univariate and multivariate data analysis considering the selected matched neonates.

Annotation	Level	m/z	Rt	FC (BPD/non-BPD)
L-Glutamic acid	1	146.0454	0.520	1.114
o-Hydroxyphenylacetic acid	1	135.0444	4.179	0.303
L-Homoserine	1	84.0450	0.587	1.136
L-beta-aspartyl-L-leucine	2	247.1296	2.126	1.332
Epigallocatechin 3'-glucuronide	3	464.0710	4.890	1.347
Physalin K	3	528.1653	3.655	1.807
Glycogen	3	689.2118	0.587	1.179
Betaine	3	118.0867	2.126	1.325
	4	231.0087	1.315	0.727
	4	357.0661	0.582	1.370
	4	277.0299	0.535	2.332
	4	438.0920	3.728	1.258
	4	872.8927	0.467	1.456
	4	277.1289	4.963	0.873
	4	185.0327	0.514	1.793
	4	322.1235	4.650	1.204
	4	142.0074	3.199	0.292

Annotation is the putative name of the compound, level is the level of annotation according to Mardegan et al18, m/z is the mass to charge ratio, Rt is the retention time in minutes and FC[BPD/non-BPD] is the fold change measured as ratio between the median of the metabolite in the BPD group and the median in the non-BPD group.

BPD, bronchopulmonary dysplasia; FC, fold change.

Figure 1. Boxplots representing the distributions of the three annotated relevant variables; black circles represent the experimental values. All relevant variables showed p-value<0.05. BPD, bronchopulmonary dysplasia.

Discussion

Gestational age is the major driver of BPD development in preterm infants with earlier gestational age associated with higher likelihood of BPD and long-term abnormalities.^{25 26} In this study, after matching for gestational age, we identified a group of urinary metabolites in the first 24 hours of life, which are related to BPD development independently from the influence of gestational age (GA) and birth weight.

Our data provide novel insights into the metabolic alterations predisposing to BPD. Metabolites such as P-Hydroxyphenylacetic acid, L-Glutamic acid, L-Homoserine are dysregulated in newborn developing BPD consistent with reduced anti-inflammatory and antioxidant mechanisms and nutritional status. Hereafter, we discuss the possible mechanisms connecting these metabolites and BPD resumed in figure 2.

Figure 2. Graphical summary of the possible mechanisms explaining experimental results. BPD, bronchopulmonary dysplasia; EUGR, extra-uterine growth restriction; 4-HPA, 4-Hydroxyphenylacetic acid; Glu, glutamate; IUGR, intra-uterine growth restriction. Created with Canva, Sydney, Australia.

L-Glutamic acid (Glu) is a non-essential α-amino acid that is used by almost all living beings in the biosynthesis of proteins but is also a precursor for the synthesis of the inhibitory gamma-aminobutyric acid a neurotransmitter. Its ionic form is known as glutamate. Glutamate also plays an important role in the body’s disposal of excess or waste nitrogen. Glutamate undergoes deamination, an oxidative reaction catalysed by glutamate dehydrogenase leading to alpha-ketoglutarate. In many respects, glutamate is a key molecule in cellular metabolism. Glu molecule offers a number of features/properties not shared by its homologs (amino adipic and aspartic acids). These properties make Glu a favourable choice for facilitating its involvement in multiple metabolic processes that play major roles in the nitrogen economy of the host, as well as serving as a nutrient, an energy-yielding substrate, a structural determinant and an excitatory molecule.²⁷ This metabolite has been shown to induce oxidative stress by binding NMDA and non-NMDA receptors, by the cysteine transporter and also by itself independently of its excitotoxic action.²⁸ Glu appeared as increased in those preterm that subsequently developed BPD, as described in low dietary protein intake,²⁹ and nutritional deficit in the prenatal (intra-uterine growth restriction) and postnatal (extra-uterine growth restriction) phases are known risk factors for BPD development. As recently described, glutamate was found as significantly increased in tracheal aspirates¹¹ and dried blood spots¹³ and our work aligns with these findings also in urinary samples, supporting the hypothesis suggested by Guo et al¹⁶ about the possible key role of glutamate in BPD prediction.

P-Hydroxyphenylacetic acid, also known as 4-Hydroxyphenylacetic acid (4-HPA), belongs to the compounds of phenolic acids that are aromatic acids that contain a phenol ring and at least one organic carboxylic acid group.³⁰ P-Hydroxyphenylacetic acid is also a microbial metabolite produced by Acinetobacter, Clostridium, Klebsiella, Pseudomonas and Proteus. Higher levels of this metabolite are associated with an overgrowth of small intestinal bacteria from Clostridia species.³¹ In lung tissues, 4-HPA attenuated hypoxia, inflammation, vascular leak and oedema. In primary rat alveolar epithelial cells, 4-HPA decreased hypertonicity-induced and hypoxia-induced (HIF)−1α protein levels, which causes lowering inflammatory cytokine levels (tumor necrosis factor-α, IL-1β and IL-6) and monolayer permeability.³² In our sample, infants with predisposition to BPD have lower levels of 4-HPA. Given the protective effects, it exerts on the lung, its reduction can be explained as a reduced ability of these infants to respond to the pathogenic noxae causing lung damage. 4-HPA also originates from the intestinal microbial flora, suggesting a possible intestinal dysbiosis in these newborns. Recently, in fact, the role of microbiome alterations in the pathogenesis of BPD is emerging in the literature.33,35 4-HPA was identified as significant both in the initial evaluation not corrected for gestational age and in the case–control evaluation, suggesting how 4-HPA levels are directly correlated with gestational age, giving an increased protection against hypoxia, inflammation, vascular leak and oedema to older preterm infants but appears to be more reduced in those infants that lately developed BPD compared with controls, suggesting also a direct involvement in the pathogenesis of the disease.

L-Homoserine is a non-essential chiral amino acid and the precursor of L-threonine and L-methionine that in our work appeared as increased in infants that developed BPD. It has been shown to be the result of the activation of adenosylmethionine hydrolase producing a-amino-gamma-butyrolactone, an unstable compound that is converted to homoserine. This enzyme is triggered in the liver by increased levels of S-adenosylmethionine (SAM), consequently higher levels of SAM or its increased metabolism can trigger higher levels of Homoserine.³⁶ SAM is considered to be the main methyl donor reagent for significant methylation reactions that occur in all living organisms, which are essential in cell differentiation and survival by regulating key metabolic pathways.³⁷ SAM is an important metabolite and can act as a nutrition, energy and stress sensor in vivo and in vitro; moreover, it is a limiting factor of the synthesis of the endogenous antioxidant glutathione.³⁸ Moreover, SAM seems to modulate the expression of antioxidant enzymes like superoxide dismutase, catalase and thioredoxin.³⁹ Consequently, we can hypothesise that L-Homoserine is an indirect marker of increased SAM as a reaction to oxidative stress or of an increased metabolism of SAM consequently reducing the defences against oxidative stress.

An aspect that shall be considered is the generalisability of our results, which were obtained in a level IV NICU of a high-income country with a BPD rate comparable to that of similar units,⁴⁰ with a matched approach reducing significantly the influence of confounders. We can consequently hypothesise a good applicability in similar settings, but their use in other contexts (eg, low income countries), where other pathogenetic mechanisms could prevail, needs to be confirmed. Our study has some other strengths like the high numerosity, the non-invasive approach (using urine) with a strict standardisation of sample collection protocols, the early timepoint evaluation and, the most important, the matched case–control approach excluding the influence of GA and other confounders.

The design of our study, with an early timepoint for sampling, is a key aspect reducing the influence of later multiple confounders appearing through the care of such a fragile population. However, this is obviously also an obstacle to the easy application of our technology requiring a (not always easy) early sample collection and a strict processing protocol. Anyhow, a later metabolic evaluation could be influenced by many other aspects of the care, acting as multiple confounders and probably reducing the predictive capability of the identified metabolites through time. This would eventually require a more complex matching procedure to confirm their predictive capability later on, or identify later strong markers of the disease. Consequently, our results can be applied only on early sampled urines.

We can identify the monocentric design as a possible weakness of the study. Indeed, the perspective of this study is a validation study in a larger, multicentric population by a target analysis. This could confirm our results and suggest an early research of these metabolites in high-risk preterm infants to nominate these subjects to innovative and known treatments at an early stage of disease development, preventing late sequelae and also reducing the risk of treating infants that do not need it.^{5 41}

Another aspect of the study is the possible validation of similar metabolites on plasmatic samples, this approach, even if invasive, would allow a faster identification of BPD developing subjects, with a consequent even earlier preventive approach. A point of care approach for these metabolites could also be developed, with the target of a fast answer to the clinicians’ demands.

The following development of our metabolomic approach will possibly be the development of drugs specifically targeting the identified pathways and consequently avoid undesired effects of non-specific treatments. Ideally, these drugs would have an anti-inflammatory effect increasing directly or indirectly P-Hydroxyphenylacetic acid, and/or a anti-oxidative target reducing the metabolism of SAM and consequently L-Homoserine levels.

Another future perspective could be the application of an integrative approach using additional proteomics, transcriptomics and genomics readout to enhance the robustness and the interpretability of the results.

Conclusions

In conclusion, our study revealed a distinctive urinary metabolome in those very preterm infants who lately developed BPD. In particular, glutamate appears as a promising metabolite in BPD prediction. The identified metabolites are markers of oxidative stress and inflammation, known aspects of BPD pathogenesis. Consequently, urine metabolomics allows a better understanding of this complex disease. Moreover, this approach appears as a promising tool to early identify preterm infants that would develop BPD and therefore apply a highly targeted preventive approach to the population at strong risk and possibly, in the future, allow the development and use of targeted drugs.

Data availability statement

Data are available upon reasonable request.