Perturbation of gene expression of the chromatin remodeling pathway in premature newborns at risk for bronchopulmonary dysplasia

The expression profiles of umbilical cords from premature newborns reveal distinct patterns, including changes in the expression of chromatin remodeling factors, associated with the development of bronchopulmonary dysplasia.


Background
The lung disorder bronchopulmonary dysplasia (BPD) occurs in 20% to 40% of infants born at under 1,000 g and before 28 completed weeks of gestation [1][2][3][4] and it is the second lead-ing cause of death among infants born within this gestational age [5]. Previously identified prenatal factors that are associated with the development of BPD include surfactant deficiency and maternal infection, including chorioamnionitis [6]. Among the multiple postnatal environmental factors and neonatal conditions that have been linked to the development of BPD are supplemental oxygen exposure, barotrauma from ventilation, the presence of patent ductus arteriosus, and neonatal infection.
Little attention has been directed to the genetics of BPD risk or the physiologic pathways, measurable at the time of delivery, that predispose preterm infants to BPD. Inflammatory mediators, including cytokines and growth factors, appear to be involved in the development of BPD [7], and their activity can be upregulated by inflammatory processes that might begin prenatally and affect the fetus via the placenta [6,8]. For example, chorioamnionitis and funisitis are accompanied by alterations in the expression of inflammatory mediators [9].
The placenta and umbilical cord tissue appear both to reflect and influence the intrauterine environment. What remains unclear is the extent to which biologic markers obtained from the umbilical cord might have clinical and developmental correlates in the fetus in both the prenatal and perinatal periods. Given the multiple tissue types in the umbilical cord that are also present in the placenta [10][11][12], we hypothesize that the umbilical cord genomic signatures may relate to the overall physiologic state of the maternal-infant unit. In this study we have explored this hypothesis by examining the relationships between human gene expression in the umbilical cord at the time of birth and subsequent BPD, defined as the requirement for supplemental oxygen at 36 weeks postmenstrual age. This also allowed us to identify characteristics of the preterm infant at risk for developing BPD.

Results
Of the 72 umbilical cord samples collected for this study, 54 contained at least the minimum of 7 µg RNA required for hybridization (13 did not) and were from infants who survived to 36 weeks postmenstrual age (5 died), and therefore at risk for BPD. All expression data are available in the National Center for Biotechnology Information's Gene Expression Omnibus repository (GSE8586).
The infants included in this study were slightly more mature than those in the Extremely Low Gestational Age Newborn (ELGAN) study [13] as a whole (Table 1). This may explain why they had a higher survival rate and lower incidences of BPD and retinopathy of prematurity.
Infants with and without BPD exhibited minimal differences in maternal characteristics, including cause of delivery, antenatal steroid exposure, race, or histopathologic evidence of chorioamnionitis (Table 2). Infants who developed BPD were of lower birth weight and less mature gestational age, required more days of supplemental oxygen and ventilation, and had higher rates of retinopathy of prematurity. They did not differ in sex ratio or rates of patent ductus arteriosus, and had a modestly higher rate of sepsis (Table 3).
Unsupervised learning revealed inhomogeneous clustering of the sample by gestational age (27 to 28 weeks versus < 27 weeks; data not shown) and by presence of BPD (BPD versus control; Figure 1). Similarly, principal components analysis of the 54 samples across the 54,675 measured transcripts revealed partial overlap of the samples across the two groups ( Figure 2). Single gene differential gene expression analysis revealed no genes with a q value (a measure of false discovery rate [FDR] [14]) below 0.05 or an uncorrected P value below 10-6. The high FDR confirmed that the overlap between the two classes of infants was high; therefore, we explored systematic differences in pathways rather than individual genes, as others have done with similarly subtle differences across human samples [15].
The gene pathways most differentially expressed by gestational age category were those related to oxidative phosphorylation, mitochondrial energy metabolism, and DNA repair ( Table 4). None of the genes within these pathways by themselves were significantly differentially expressed, but the q value for the entire set of genes within each pathway (of the more than 600 sets evaluated) was below 0.0001, as calculated using the sigPathway package [16].
When the samples were evaluated comparing infants with BPD versus those without BPD, the most differentially expressed gene pathways included the aforementioned bioenergetic pathways, as well as histone acetyltransferase binding activity and chromatin remodeling pathways (Table  5). Although the individual genes within these pathways were not significantly differentially expressed, the expression of the entire 'chromatin packaging' pathway relative to the overall transcriptome was highly significantly differentially expressed ( Figure 3). The ten most differentially expressed genes of those in chromatin remodeling pathways are shown in Figure 4 to illustrate how this finding is only a trend at the individual gene level.

Discussion
Microarray profiling has been used to classify and predict human disease, but this technique has not yet been applied widely to the investigation of diseases of the premature new-born infant. Expression analysis has proved informative in murine models of fetal and postnatal lung injury [17]. Neonatal lung samples, however, are not routinely available. Umbilical cords are routinely available, and their availability has allowed us to explore expression profiles at birth with respect to gestational age and subsequent development of BPD. This has provided a rare opportunity to examine the influence of fetal physiology on postnatal health and development using the multiple tissues in umbilical cords as a proxy for a wide variety of tissues in the maternal-fetal unit.
Gene expression signatures alone provide imperfect clustering of different gestational age groups in the unsupervised analyses. For this reason, as expected, the single gene differential expression measures were fraught with high false discovery rates. However, the supervised analysis at the level of entire pathways [15] revealed highly significant distinctions for both gestational age and for the subsequent development of BPD.
Several biologic pathways have emerged from this investiga- Table 2 Baseline characteristics of mothers and placentas related to infant outcomes in terms of BPD  tion, characterizing different early physiologic states: bioenergetics (Krebs cycle, mitochondrial function, and oxidative phosphorylation), transporter activity, DNA synthesis and repair, and chromatin remodeling. However, the latter pathway was consistently present at a frequency greater than expected by chance alone (it was enriched) in the comparisons of all infants with BPD compared with control infants. These findings are consistent with the earlier referenced single gene investigations of prematurity implicating inflammatory and bioenergetic processes [6][7][8][9]18].

Chromatin remodeling
The chromatin remodeling apparatus is involved in the inflammatory pathways in adult lung disease [19]. Indeed, glucocorticoids (one of the mainstays of therapy for pulmonary diseases) reverses histone acetylation of activated inflammatory genes by binding liganded glucocorticoid receptors to coactivators, by recruiting histone deacetylase-2 to the activated transcription complex [19][20][21]. However, to our knowledge, this is the first documentation in human preterm neonates of the relative activity of the histone acetylation/chromatin remodeling pathway at birth in individuals who subsequently develop BPD. This intersection with the pathophysiology of chronic obstructive pulmonary disease, although very limited, does raise the question of whether a subset of infants with BPD and adults with chronic obstructive pulmonary disease share a common vulnerability that is exposed by different stressors. Such commonality might have practical import for prevention and treatment [21].
In our study, the proportions of infants exposed to antenatal glucocorticoids were similar in infants who did and those who did not develop BPD. Thus, differential administration of corticosteroids did not distort the relationship between gene expression and development of BPD.

Bioenergetic pathways
Differentially expressed energy pathways seen in younger premature infants when compared with older premature infants suggest that the timing of delivery may have a global effect on energy metabolism. Although human data are lacking, preterm rats tend to have diminished mitochondrial content and function compared with term born rats [22]. The sets of genes characteristic of oxidative phosphorylation and other bioenergetics were expressed at lower levels in BPD infants than in control infants and are also developmentally regulated, with lower levels of expression in the least mature. Consequently, in this small sample we could not separate a gestational age effect from a BPD propensity.
In summary, this study of RNA expression profiles in umbilical cord tissue demonstrates the potential of this technology in investigations of perinatal and postnatal disorders. Because chromatin remodeling pathways appeared to be differentially regulated in umbilical cord tissues of the subsequently BPD-affected neonates, therapeutic modalities that are being explored for treatment of adult pulmonary diseases with similar molecular pathophysiology (for example, steroids with fewer side effects such as RU24858 and RU40066 [21] and antioxidants [23,24]) may hold considerable new promise for those infants who are at risk for BPD. The analytic methodologies used here might also be sufficiently robust to identify individuals who are at risk. Although entire pathways were significantly differentially regulated between BPD and control infants, multigenic predictors based on these pathways did not exhibit strong performance, and identification of more predictive biomarkers will require larger premature neonate sample sizes

Materials and methods
The study population consisted of a subset of preterm infants born at one of three centers (Brigham and Women's Hospital, Beth Israel Deaconess Medical Center, and Wake-Forest Table 4 Differentially expressed gene sets for gestational age under 27 weeks versus 27 to 28 weeks Gene set/pathway NT k q value NE k q value Shown here are the pathways (with redundant pathways removed) that were ranked at the top of the list of those differentially expressed by two measures of false discovery: the NT k q value and the NE k q value. The NT k q value corresponds to the degree to which genes within a set/pathway are more predictive of the phenotype than the genes outside that set, and the NE k q value corresponds to the degree to which genes within that set are predictive of the phenotype. Only those pathways with an NT k q value below 10 -4 and a NE k q value below 1 are shown above. The relatively high NE k q value with respect to NT k q value values shows that these gene sets are not as significantly predictive of outcome, because they demonstrate significant differential expression with respect to the other gene sets. This illustrates the limitations of the sample size and nature of the current data set. Each pathway is annotated by ontology source (Gene Ontology [GO]  Unsupervised clustering based on Euclidean distance in expression between samples Figure 1 Unsupervised clustering based on Euclidean distance in expression between samples. Each row corresponds to a gene and each column (labeled at the bottom) corresponds to an infant who subsequently developed bronchopulmonary dysplasia (BPD) or a control infant.
Total RNA was isolated from umbilical cord tissue homogenates. Tissue samples were homogenized in 1 ml TRIZOL reagent using a power homogenizer. Chloroform (0.2 ml) was then added. Samples were centrifuged at no more than 12,000 g for 15 min at 2 to 8°C. RNA was precipitated from the aqueous phase by mixing with 0.5 ml isopropyl alcohol and centrifuged at no more than 12,000 g for 10 min at 2 to 8°C. RNA pellet was washed with 1 ml of 75% ethanol. RNA was dissolved in 100 µl RNase-free water. The solution was re-suspended in 100 µl water and incubated at 37°C for 5 min. Qiagen RNeasy Mini Kit (Qiagen Inc. Valencia, CA, USA) along with 350 µl Buffer RLT (with B-mercaptethanol) was added to 100 µl dissolved RNA. Then, 250 µl ethanol (96% to 100%) was added to the diluted RNA. The sample (700 µl) was applied to an RNeasy mini column placed in a 2 ml collection tube, and then centrifuged at ≥ 8,000 g (≥ 10,000 rpm). buffer RPE (500 µl) was pipetted onto the RNeasy column and then centrifuged at ≥ 8,000 g (≥ 10,000 rpm). Northern blot analysis was performed on each sample to confirm quality. AffymetrixAE U133 chips (Affymetric Inc., Santa Clara, CA, USA) were used for hybridization.
Demographic and clinical information was gathered from patient charts, ELGAN clinical data collection sheets, and maternal interviews. The primary outcome of BPD was defined as a persistent oxygen requirement at 36 weeks postmenstrual age.

Computational analysis
Samples were evaluated by outcome of BPD and by gestational age. Single gene analysis was performed using the multtest and qvalue packages from Bioconductor [25] to identify those genes that were differentially expressed by the t test and ranked by FDR.
Gene set enrichment analysis was conducted using the Sigpathway package of Tian and coworkers [16] using 1,000 permutations for each analysis. Only the top ranked pathways with an NT k q value below 10 -4 and an NE k q value below 1.0 were included in the results. The NT k q value corresponds to the degree to which genes within a set/pathway are more predictive of the phenotype than the genes outside that set, and the NE k q value corresponds to the degree that genes within that set are predictive of the phenotype.
Principal component analysis was performed using the R princomp function. Of the measured variance in the transcriptome, 98.5% was captured in the first two principal com-ponents; therefore, the plot in Figure 2 only includes the first two components. To demonstrate the extent of the overlap, an ellipsoid hull for each class (BPD and control infants) was calculated as the ellipsoid of minimal area, such that all given points lie just inside or on the boundary of the ellipsoid [26]. Table 5 Differentially

expressed pathways in infants with BPD versus infants without BPD
Gene set/pathway NT k q value NE k q value Shown here are the pathways (with redundant pathways removed as well as those that overlapped with Table 4) that were ranked at the top of the list of those differentially expressed by two measures of false discovery: the NT k q value and the NE k q value. The NT k q value corresponds to the degree to which genes within a set/pathway are more predictive of the phenotype than the genes outside that set, and the NE k q value corresponds to the degree to which genes within that set are predictive of the phenotype. Only those pathways with an NT k q value below 10 -4 and a NE k q value below 1 are shown above. Each pathway is annotated by ontology source (Gene Ontology [GO] or humanpaths). BPD, bronchopulmonary dysplasia; PI3K, phosphoinositide-3 kinase.
Relative expression of DNA packaging gene set relative to the overall transcriptome

BPD/Control
An outline was then drawn of each corresponding ellipsoid in Figure 2. The two P values on a t test on the samples in each of the first two principal components were greater than 0.05, confirming the visual impression.
Hierarchical clustering (using the R hclust function) based on Euclidean distance of expression between samples was per-formed using all those genes of the 54,675 that had at least one value greater than the mean in the 54 samples.
Box-plots of differential expression of the ten most differentially expressed genes in SWI/SNF chromatin remodeling pathway Figure 4 Box-plots of differential expression of the ten most differentially expressed genes in SWI/SNF chromatin remodeling pathway. Individually, none of these genes reached significance for differential expression, although the pathway was highly significantly enriched.