Cytokine and growth factor correlation networks associated with morbidities in extremely preterm infants

Abstract

Background: Cytokines and growth factors (GF) have been implicated in the development of retinopathy of prematurity (ROP) and bronchopulmonary dysplasia (BPD). We hypothesize that even small coordinated changes in inflammatory proteins or GFs may reveal changes in underlying regulating mechanisms that do not induce obvious changes in concentration of individual proteins. We therefore applied correlation network analysis of serum factors to determine early characteristics of these conditions.

Methods: Concentrations of 17 cytokines and five GFs were measured and analysed in blood samples from cord blood, on day one and during the following month in 72 extremely preterm infants. Spearman's correlation networks distinguishing BPD and severe ROP patients from non-affected were created.

Results: Most cytokine concentrations correlated positively with each other and negatively with GFs. Very few individual cytokines differed between patients with and without ROP or BPD. However, networks of differently correlated serum factors were characteristic of the diseases and changed with time. In ROP networks, EPO, G-CSF and IL-8 (cord blood), BDNF and VEGF-A (first month) were prominent. In BPD networks, IL-1β, IGF-1 and IL-17 (day one) were noted.

Conclusions: Network analysis identifies protein signatures related to ROP or BPD in extremely preterm infants. The identified interactions between serum factors are not evident from the analysis of their individual levels, but may reveal underlying pathophysiological mechanisms in the development of these diseases.

Keywords: Bronchopulmonary dysplasia; Correlation network analysis; Cytokine interactions; Extremely preterm infants; Retinopathy of prematurity.

Fig. 1

Correlations between serum factors in ROP. Bubble plots representing correlations between the levels of cytokines and GFs in cord blood (A—C), in peripheral blood within the first day (D—F) and the first month after birth (G—I) for patients with and without ROP. Plots in A, D, and G show correlations in the groups with severe ROP, and B, E, and H correlation in the groups without ROP. Red represents positive and blue negative correlations with deeper color for larger absolute value of correlation coefficient. The bubble size indicates significance of difference (P values for individual correlations in the lower part of bubble plot). Plots C, F, and I show the difference between severe ROP and no ROP for the corresponding period (P < 0.05). Correlations are based on residuals after correction for gestational age

Fig. 2

Heat maps and networks for cytokines and GFs in severe ROP. Panel A shows heat maps where serum factors have been ranked (graded by color) by the sum of the z-scores of their correlations. The number of significant correlations for the factor given after the colon. Panel B shows a similar ranking of the differences in correlations between patients with severe ROP and without ROP. Panel C shows ROP-related networks of factors with correlations different between severe ROP and no ROP. Red indicates positive, blue negative, grey non-significant correlations. Line thickness reflects correlation strength. GFs are labeled green

Fig. 3

ROP-related networks for serum factors in patients with and without various diagnoses. Networks identified for severe ROP (Fig. 2) were analyzed in patients grouped by exposure to intrauterine inflammation (histological chorioamnionitis (HCA) and/or fetal inflammation (FIRS) based on placenta examinations, panel A) or fetal growth (birth weight standard deviation score (BW-SDS) panel B) at different time points. Same marking of networks as in Fig. 2 C. Asterisks on yellow background show significant differences between patients with and without the respective condition

Fig. 4

Correlations between serum factors in BPD. Bubble plots representing correlations between levels of serum factors (cytokines, GFs) in cord blood (A—C), in peripheral blood within the first day (D—F), and during the first month after birth (G—I) for patients with or without BPD. Factors significantly different between groups are presented in (C, F) and (I) plots (P < 0.05). Values are corrected for gestational age. See Fig. 1 for a detailed description

Fig. 5

Heat maps and networks for cytokines and GFs in BPD. Panel (A) shows heat maps in patients with or without BPD; panel (B) shows a similar ranking of the differences in correlations between these patients. Panel (C) shows BPD-related networks—same presentation as in Fig. 2

Fig. 6

BPD-related networks for serum factors in patients with and without various diagnoses. Networks identified for BPD (Fig. 5) were analyzed in patients grouped by presence of placental and/or fetal inflammation (FIRS or HCA, panel A) or bodyweight deviation (BW-SDS, panel B) at different time points. Same marking of networks as in Fig. 2C. Asterisks on yellow background show significant differences between patients with and without the respective condition