Baseline Characteristics
This study involved 347 very preterm infants, with 170 in the BPD group and 177 in the control group. Both groups exhibited similar baseline characteristics, including gestational age, length of rupture of membranes, clinical chorioamnionitis, gestational diabetes, antenatal corticosteroids, surfactant administration, and caffeine administration. However, infants in the BPD group had lower birth weight and Apgar Scores at 5 minutes compared to the control group. Additionally, the BPD group had higher rates of cesarean section, hypertension or pre-eclampsia, and sepsis than the control group (all p<0.05). Furthermore, the BPD group required longer durations of mechanical ventilation, continuous positive airway pressure (CPAP), and oxygen supplementation compared to the control group (all p<0.001) (Table 1).
Metabolites Analysis
OPLS-DA Model Building and Validation
The OPLS-DA score plot depicted clear clustering between very preterm infants with and without BPD (Fig. 1 A). The OPLS-DA model exhibited R2Xcum (0.710) values greater than 0.5, indicating a strong fit of the model to the data. The first two principal components accounted for over 50% of the variation in the metabolites analyzed, further supporting the model's efficacy. Additionally, permutation analysis (Fig. 1B) confirmed the model's validity, as all permuted R2s and Q2s values were lower than the original values, suggesting that the model was not randomly generated.
Contribution Analysis of All Metabolites to BPD
Using a contribution plot, metabolites were ranked based on their contributions to the model (Fig. 1C). Fourteen metabolites were identified as potential discriminant metabolites for disease prediction (VIP > 1.0): methylglutarylcarnitine (C6DC) (VIP = 2.519), 3-hydroxylpalmitoylcarnitine (C16OH) (VIP = 2.174), Methionine (Met) (VIP = 1.957), Alanine (Ala) (VIP = 1.868), free carnitine (C0) (VIP = 1.686), isovalerylcarnitine (C5) (VIP = 1.650), glutarylcarnitine (C5DC) (VIP = 1.614), Leucine (Leu) (VIP = 1.461), butyrylcarnitine (C4) (VIP = 1.392), acetylcarnitine (C2) (VIP = 1.371), oleylcarnitine (C14OH) (VIP = 1.230), linoleoyl carnitine (C18:2) (VIP = 1.072), Ornithine (Orn) (VIP = 1.056), and Tyrosine (Tyr) (VIP = 1.054). S-plots indicated significant differences in C6DC, C16OH, Met, and Ala levels between the two groups. C6DC and C16OH showed a positive correlation, while Met and Ala exhibited a negative correlation (Fig. 1D).
Univariate Analysis for Cross-Validation
The univariate analysis of all metabolites (Tables 2 and 3) was complemented by multivariate analysis using OPLS-DA to identify changes in metabolites associated with the development of BPD. The top 14 potential discriminant metabolites were highlighted. The distribution pattern of each metabolite was evaluated to assess the significance of the tests.
The scatter plot with bars illustrated a higher C6DC ratio in the BPD group compared to the control group (p<0.001). Met, Ala, Leu, C0, and C2 levels were lower in the BPD group than in the control group (all p<0.05), while C6DC, C16OH, C5, C5DC, and C4 levels were higher in the BPD group (all p<0.05) (Fig. 2). Multivariate logistic regression analyses indicated that changes in C6DC levels remained significant even after adjusting for factors such as birth weight, Apgar Scores, maternal hypertension, cesarean section, duration of Mechanical Ventilation, CPAP, and Oxygen, suggesting that C6DC is an independent risk factor for BPD.
The predictive performance of potential metabolic biomarkers for BPD was evaluated using an ROC plot. The AUC values for all identified metabolites were > 0.5, with C6DC and C16OH having significantly higher AUC values (p=0.001 and p=0.003, respectively), indicating good predictive ability. The sensitivity and specificity were 51.8% and 63.3%, respectively, for C6DC, and 66.5% and 46.3%, respectively, for C16OH, suggesting that C6DC and C16OH could serve as potential biomarkers for diagnosing BPD.
Differential analysis of metabolites
The study identified 14 metabolites in both the BPD and control groups, which were selected based on criteria of p<0.05 or VIP > 1 for further evaluation of their impact on BPD. Analysis of correlation heat maps revealed significant negative correlations of Ala, Met, Orn, Tyr, C2, and C5 with C6DC, a positive correlation of C5DC and C2 with C0, and a positive correlation of C0, C2, and C4 with C5DC. The Mantel test indicated that the severity of BPD was associated with C0, C2, C4, and C5DC, while BNP levels were related to C0. (Fig. 3)
Kyoto Encyclopedia of Genes and Genomes enrichment analysis suggested that metabolites such as Ala, Leu, Met, Orn, C0, C18, C4, and C5DC were involved in metabolic pathways, secondary metabolites biosynthesis, D-Amino acid metabolism pathway, adenosine triphosphate-binding cassette transporter pathway, and amino acid pathway biosynthesis. (Fig. 4)