Network analysis of temporal effects of intermittent and sustained hypoxia on rat lungs

Abstract

The molecular networks underlying the lung response to hypoxia are not fully understood. We employed systems biology approaches to study temporal effects of intermittent or sustained hypoxia on gene expression in rat lungs. We obtained gene expression profiles from rats exposed to intermittent or sustained hypoxia lasting 0–30 days and identified differentially expressed genes, their patterns, biological processes, and regulatory networks critical for lung response to intermittent or sustained hypoxia. We validated selected genes with quantitative real-time PCR. Intermittent and sustained hypoxia induced two distinct sets of genes in rat lungs that displayed different temporal expression patterns. Intermittent hypoxia induced genes mostly involved in ion transport and homeostasis, neurological processes, and steroid hormone receptor activity, while sustained hypoxia induced genes principally participating in immune responses. The intermittent hypoxia-activated network suggested a role for cross talk between estrogen receptor 1 (ESR1) and other key proteins in hypoxic responses. The sustained hypoxia-activated network was indicative of vascular remodeling and pulmonary hypertension. We confirmed the temporal expression changes of 12 genes (including the Esr1 gene and 4 ESR1 target genes) in intermittent hypoxia and 8 genes in sustained hypoxia with quantitative real-time PCR. Conclusions: intermittent and sustained hypoxia induced distinct gene expression patterns in rat lungs. The functional characteristics of genes activated by these two distinct perturbations suggest their roles in the downstream physiological effects of intermittent and sustained hypoxia. Our results demonstrate the discovery potential of applying systems biology approaches to the understanding of mechanisms underlying hypoxic lung response.

hypoxia has profound physiological effects. Traditionally, studies examining pathophysiological effects of hypoxia have mainly focused on sustained (or continuous) hypoxia (SH), particularly in the context of high-altitude physiology. More recently, with the increased interest in sleep-disordered breathing, the physiological effects of intermittent hypoxia (IH) have also been studied (27), and response differences between these two hypoxic paradigms have been outlined.

In the clinical setting, IH and SH are found in distinctly different conditions. IH is primarily associated with sleep-disordered breathing, which has been increasingly recognized as a highly prevalent and clinically important set of disorders (25). In contrast, SH is associated with chronic obstructive pulmonary disease, pulmonary fibrosis, and other parenchymal lung disorders that adversely affect alveolar gas exchange (16). The morbid consequences of these two conditions also differ: while IH is mainly associated with systemic hypertension, cardiovascular morbidity and mortality, and neuronal and humoral disorders, long-term SH most commonly leads to pulmonary hypertension and right ventricular (RV) failure (10, 26, 28, 29, 32, 33).

In physiological terms, IH and SH exert somewhat different effects on the vascular, cardiac, and neurological systems. IH induces systemic arterial blood pressure (BP), increased sympathetic nervous activity, and respiratory long-term facilitation (15, 20), while SH may induce pulmonary vascular remodeling (e.g., vascular wall thickening) and elevated pulmonary artery pressures, without evidence for altered systemic BP (13, 32). Both IH and SH can enhance the hypoxic ventilatory response; however, SH appears to induce higher early ventilatory increase and more rapid ventilatory decline than IH (27).

At the molecular level, IH and SH can both induce a variety of responses in animals, which is believed to be modulated mainly by hypoxia-inducible factor (HIF)-1. HIF-1 can induce hypoxic responses to increase O₂ delivery and energy production, which is critical for animals to survive in hypoxia (1, 31, 40). Similar responses have also been implicated in pathophysiological mechanisms of vascular remodeling and pulmonary hypertension observed in patients suffering from SH (35). Despite the important role of HIF-1 in hypoxia, the hypoxic effects elicited by HIF-1 alone cannot fully explain physiological abnormalities occurring in patients suffering from IH, suggesting that other regulators may play key roles in IH. A majority of studies on hypoxic effects in animals have examined the response of the brain, carotid body, and cardiovascular system to IH or SH.

In this study, we employed gene expression profiling to identify temporal expression patterns of genes in rat lungs in response to either IH or SH. Systems biology approaches were applied to globally characterize distinct molecular programs and potential regulatory networks that play important roles in IH and SH.

MATERIALS AND METHODS

Experimental Procedures

Adult inbred Sprague-Dawley male rats (weights 250–275 g) were placed in oxygen- and carbon dioxide-controlled chambers. Rats were subjected to either IH or SH exposure as previously described for 1–30 days of duration (11, 27). Briefly, rats subjected to the IH treatment were exposed to 90-s cycles of 10% inspired oxygen followed by 90-s of 21% inspired oxygen for 12 h of simulated daylight, when rats sleep; for the remaining 12 h during the night, the rats were exposed to 21% inspired oxygen (11). Rats subjected to the SH treatment were exposed to 10% inspired oxygen at all times. Carbon dioxide concentrations were monitored and maintained at <0.01% by altering chamber ventilation (27). All experimental protocols were approved by the Institutional Animal Use and Care Committee at the University of Louisville and were in accordance with National Institutes of Health requirements for the care and use of laboratory animals.

Measurement of Systemic Blood Pressure

Animals (n = 6) were placed into a Plexiglas cylindrical restrainer with a size-appropriate cuff around the tail. Automated occlusion plethysmography (tail cuff method) was used, which involved an analog-to-digital signal converter and a digital acquisition and analysis system (Kent Scientific, Torrington, CT). During a 10-min accommodation period, body temperature was raised to 38–39°C with a heating pad, so as to increase blood flow in the tail. Heating was discontinued once the BP signal was detectable. At least 8–10 BP readings were obtained from each rat at each time point, and all measurements were conducted at the same time of day, namely, 10:00 AM; these values were then averaged to yield the mean BP for that time point for each rat. Systolic and diastolic pressure were measured from six rats before initiation of hypoxic exposures, and then at the designated time points for up to 30 days. The time course for BP changes throughout the experimental period was plotted for each experimental group.

Telemetric Measurement of Right Ventricular Pressure

RV blood pressure (RVP) in rats (n = 6) was measured with an implanted radiotelemetry system (Dataquest A.R.T. 2.1; Data Sciences). The transmitter (model TA11PA) was connected to a fluid-filled sensing catheter and transferred the signals to a remote receiver (model RPC-1) and a data-exchange matrix connected to a computer. Under intraperitoneal ketamine and xylazine anesthesia, the sensing catheter was inserted into the jugular vein and forwarded to the RV. The waveform was displayed on the computer and used to ensure correct positioning of the catheter. Animals were allowed to recover and were housed individually in standard rat cages. RVPs were recorded at 1-h intervals from the time of implantation and are presented as the mean RVPs averaged over the 24-h period.

Tissue Acquisition and RNA Extraction

Rats were exposed to either IH or SH and were killed after 1, 3, 7, 14, or 30 days. Rats exposed to room air (RA) served as controls. Rats were anesthetized with a pentobarbital overdose (100 mg/kg ip) and euthanized by decapitation. Whole lungs were then rapidly removed, snap frozen over liquid nitrogen, and stored at −70°C until needed for further use. Total RNA was extracted as previously described (24) and used for microarray experiments and quantitative real-time polymerase chain reaction (qRT-PCR).

Microarray Protocol

Microarray experiments were performed with a protocol recommended by the manufacturer of the arrays and as previously described by us (43). Briefly, extracted RNA from each rat was used as a template to generate cRNA, which was then labeled with a fluorescent dye and hybridized to a CodeLink UniSet Rat I Bioarray presynthesized with oligonucleotide probes (Applied Microarrays). Microarrays were washed and scanned as recommended.

Quantitative Real-Time PCR

The detailed protocol for qRT-PCR can be found in Ref. 24, with modification. Briefly, cDNA was synthesized from 2 μg of total RNA. This was performed with SuperScript First-Strand Synthesis System (Invitrogen, 11904-018). The qRT-PCR was performed in a 384-well format, under standard cycling conditions, with reagents, primers, and probes for TaqMan Gene Expression Assays (Applied Biosystems). Finally, the ABI Prism 7900HT instrument and Sequence Detection Software v2.2 (SDS 2.2) (Applied Biosystems) were used for PCR quantification and measurement of the PCR amplification product. Gene-specific primers and probes used were as follows. Estrogen receptor 1 (Esr1): Rn00562166_m1; Potassium inwardly rectifying channel, subfamily J, member 6 (Kcnj6): Rn00755103_m1; Purinergic receptor P2X, ligand-gated ion channel, 2 (P2rx2): Rn00586491_m1; Mitogen-activated protein kinase 1 (Mapk1): Rn00671828_m1; Amphiregulin (Areg): Rn00567471_m1; L-type voltage-dependent calcium channel, α1C subunit (Cacna1c): Rn00709288_m1; Nuclear receptor subfamily 0, group B, member 1 (Nr0b1): Rn00584062_m1; Cytochrome P-450, family 26, subfamily b, polypeptide 1 (Cyp26b1): Rn00710377_m1; Cytochrome P-450, family 2, subfamily b, polypeptide 15 (Cyp2b15): Rn00755182_g1; ATPase, Na⁺/K⁺ transporting, α2 polypeptide (Atp1a2): Rn00688124_gH; Bone morphogenic protein 2 (Bmp2): Rn00567818_m1; MAD homolog 1 (Drosophila) (Smad1): Rn00565555_m1; Endothelin 1 (Edn1): Rn00561129_m1; Matrix metallopeptidase 9 (Mmp9): Rn01423075_g1; CD8b molecule (Cd8b): Rn580581_m1; Bone morphogenic protein receptor, type II (serine/threonine kinase) (Bmpr2): Rn01437210_m1; Glucuronidase, β: Rn00566655_m1.

Microarray Data Preprocessing

In our initial experimental design, there were three biological replicates for each time point (i.e., microarray data were obtained from RNAs extracted from three different rats). However, because of quality issues, some microarrays were eliminated before data preprocessing, and therefore there are fewer than three biological replicates for some time points (see Supplemental Table E1 for details).1 Raw microarray data were preprocessed with a protocol previously described by us (43) and also available in the Supplemental Data. The processed microarray data was log₂-transformed and normalized with a statistical method, CyclicLoess, to minimize unwanted noise in the data (43). We previously showed (43) that CyclicLoess is one of the most effective normalization methods for reducing intensity-dependent dye effects in CodeLink Bioarray data. Finally, the log₂-transformed, normalized data were used as input to the downstream computational analyses. The complete microarray data set is available at the Gene Expression Omnibus (GEO) database (http://www.ncbi.nlm.nih.gov/geo; GSE8705).

Computational Analyses

Local regression and spectral analysis.

We have developed a flexible yet rigorous nonparametric statistical procedure, local regression and spectral analysis (LRSA), which estimates genes differentially expressed at different time points with respect to the control (RA, day 0) with a nonparametric local regression smoothing model and identifies temporal expression patterns of these genes with a spectral clustering algorithm. The LRSA procedure is described in detail in Ref. 44. Briefly, LRSA is a two-step procedure. In the first step, temporal expression data for each transcript in each microarray data set are first fitted with a local polynomial quadratic (degree = 2) model within a smoothing parameter or bandwidth h_i, where h_i is selected optimally with the generalized cross-validation (GCV) score (36). We chose to use the degree of 2, instead of 1, for the local polynomial model since our results showed that the former choice led to more powerful results and yielded lower false discovery rates (FDRs) than the latter choice (data not shown). To determine whether a gene is differentially expressed over time with respect to the control time point, we calculated the simultaneous 95% confidence intervals for the fitted (or expected) intensity values at time point k with a method due to Faraway and Sun (7); this method guarantees that these confidence intervals are simultaneously valid for all the time points in the data, and it also allows the estimate of heteroscedasticity in the data (7, 36, 41). Since there was only one biological replicate at some time points in both IH and SH data, we used a modified procedure described in detail in the Supplemental Data to estimate the simultaneous 95% confidence intervals at these time points. In addition, we used external control probes on the CodeLink arrays as a surrogate to estimate the FDR associated with LRSA, and thus to guide the detection of differentially expressed genes; we call the surrogate metric “FDR for external control probes” (FDR_EC). In the Supplemental Data, we compare P values obtained at some test time point K relative to the control time point for both external control probes and regular transcripts on the arrays; it can be seen that P values for these different types of probes follow similar distributions. In the second step, LRSA clusters the differentially expressed genes resulting from the first step with a spectral clustering algorithm. A web server for identifying temporal differentially expressed genes with LRSA can be accessed at http://www.pitt.edu/∼wew16/HypoxiaNetwork/index.html. In addition, our program written in R and Matlab for the LRSA procedure as well as the results yielded by computational analyses in this work are also available from the same web site.

To identify differentially expressed genes during either IH or SH, we applied LRSA without multiple testing control to the two hypoxia data sets, respectively. We have shown that when applied to the IH data set LRSA without multiple testing control is much more powerful and yet yields the same FDR_EC as LRSA with multiple testing control (see our unpublished work in Ref. 44 and additional details in the Supplemental Data). We determined a gene as differentially expressed in the IH data set only if its expression at time point k relative to the control time point (day 0) satisfied: 1) P < 0.05 and 2) fold change ≥ 2 (on the original data scale). We showed in Ref. 44 that when these criteria were used we detected differentially expressed genes from the IH data set with high power and FDR_EC = 0. To identify differentially expressed genes from the SH data set, we employed the same criteria as we did for the IH data set on array data obtained from rats exposed to SH for 1, 7, and 30 days; we used a slightly higher fold change, fold change ≥ 2.5, to determine differentially expressed genes from the data obtained from rats with 3-day SH exposure since the data have a higher variability. Because the quality of the arrays obtained from rats exposed to SH for 14 days was not reliable, we did not include them in our analysis. Differentially expressed genes identified from the SH data set also satisfied FDR_EC = 0.

Gene Ontology statistical analysis.

To identify functional groups of genes enriched among differentially expressed genes in rat lungs in response to either IH or SH, we performed Gene Ontology (GO) analysis with the GOstat program (3). The GOstat program finds the enriched functional groups using Fisher's exact tests with the aid of the GO annotation (3). The list of the genes on the CodeLink UniSet Rat I Bioarrays was used as the reference gene list for GOstat. To determine whether a functional group is significantly enriched, we employed the following criteria: 1) the unadjusted P value < 0.005, which amounted to P < 0.1 with the FDR-controlling procedure of Benjamini and Hochberg, and 2) the number (n_i) of the hypoxia-induced genes in GO group i satisfies n_i > 10, or the percentage (p_i%) of the hypoxia-induced genes assigned to GO group i (i.e., p_i% = n_i/N_i, where N_i is the total number of the genes in GO group i) satisfies p_i > 20.

Identifying potential ESR1 target genes in rats.

To determine potential ESR1 target genes among differentially expressed genes in rat lungs during either IH or SH, we identified all such genes in the rat by using the lists of the potential human and mouse target genes of ESR1 published by Bourdeau et al. (4) and also available at http://www.mapageweb.umontreal.ca/maders/eredatabase. Specifically, we first mapped all human and mouse ESR1 target genes to their rat homolog genes; then we identified a gene as a potential ESR1 target gene in the rat only if it appears on both lists of the rat homologs of human and mouse ESR1 target genes.

Network analysis.

To identify regulatory networks activated by either IH or SH, genes differentially expressed during each type of hypoxia were analyzed with Ingenuity Pathways Analysis (IPA) software (Ingenuity Systems, http://www.ingenuity.com). We have previously employed the IPA network analysis to detect informative regulatory networks in asthma (21). The final IH- or SH-activated network was formed by merging 14 subnetworks identified by IPA that were assigned the highest scores; the network consisted of ∼500 genes, the biggest network allowed to be identified by IPA. Additional information on the procedure of the network analysis can be found in the Supplemental Data.

Disease relevance analysis of a network.

This analysis identified the physiological malfunctions significantly associated with the detected network and was performed with IPA. Fisher's exact test was used to calculate the P value determining the probability that each malfunction associated with the network is due to chance alone. We determine an association as significant if 1) P < 0.05 and 2) there are more than a cutoff number (20 or 30 was used in this work) of genes in the network involved in the association.

RESULTS

Physiological Responses of Rats to IH or SH

IH and SH elicited different physiological responses as far as systemic BP and RVP. Systemic BP was increased slightly in IH after 30 days of exposure (P < 0.05, relative to RA control rats) but remained unaltered in SH (Fig. 1A). Mean RVPs were increased in rats exposed to either IH or SH (P < 0.05 relative to RA control rats), but the increase was significantly greater in SH than in IH (Fig. 1B).

Fig. 1.

Systemic diastolic (DBP) and systolic (SBP) blood pressures and mean right ventricular blood pressure (RVP) in rats exposed to intermittent (IH) and sustained (SH) hypoxia. SBP and DBP (A) and RVP (B) were all measured in rats exposed to IH or SH for 0 [room air (RA)], 1, 3, 7, 14, and 30 days. Data are means ± SE (n = 6). Significant differences: *P < 0.05 compared between rats exposed to either IH or SH and RA control rats; ⁺P < 0.05 compared between rats exposed to IH and SH.

Comparison of Expression Patterns of Differentially Expressed Genes in IH and SH

Using the nonparametric LRSA procedure developed by us to analyze time course expression data, we detected two sets of genes differentially expressed over time in rat lungs induced by IH and SH, respectively (Supplemental Tables E2 and E5; the gene lists can also be found at http://www.pitt.edu/∼wew16/HypoxiaNetwork). These two sets of genes exhibited different expression patterns (Fig. 2, A and C) and participated in different functions as well as forming different networks as suggested by multiple computational analyses. In the following, we present gene expression patterns and functional and network analyses of the differentially expressed genes activated during IH and SH in detail.

Fig. 2.

Temporal expression profiles of genes differentially expressed in rat lungs in response to either IH or SH. A: heat map of the expression profile of differentially expressed genes in IH. d, Day; cl, cluster. B: histogram of genes differentially expressed during IH on each measured day (D). C: heat map of the expression profile of differentially expressed genes in SH. D: histogram of genes differentially expressed during SH on each measured day.

Temporal Gene Expression Patterns in IH

We identified 1,525 genes differentially expressed in rat lungs during IH with LRSA (Supplemental Table E2). These genes exhibited seven temporal expression patterns (Fig. 2A), which appeared to be homeostatic: expression of genes peaked at earlier time points and converged to the control level at later time points. An analysis of the numbers of the genes induced over time during IH revealed a bell-shaped distribution (Fig. 2B), which indicated that a vast majority of the genes were induced on day 7, while only a few genes were induced at earlier (days 1 and 3) or later (days 14 and 30) time points.

The upregulated genes in response to IH reside in clusters 1–4 (Fig. 2A and Supplemental Fig. E1A) and appeared to be induced in a time-dependent manner: genes in clusters 1 and 2 exhibited early-to-middle patterns, cluster 3 an early acute pattern, and cluster 4 a middle-to-late pattern. The downregulated genes in IH reside in clusters 5–7 (Fig. 2A and Supplemental Fig. E1A); unlike the upregulated genes, the downregulated genes appeared to have more uniform patterns.

Functional Theme Analysis of Genes Enriched in IH

GO enrichment analysis revealed a large number of functional groups of genes significantly enriched in the upregulated clusters (P < 0.1 with FDR, Fig. 3A and Supplemental Fig. E2), which contained genes involved in system development, neurological process, ion transport and homeostasis, G protein-coupled receptor signaling, behavior, oxidoreductase, and steroid hormone receptor activity; in contrast, there were few functional groups enriched in the downregulated clusters, which included genes participating in nucleic acid binding and ribonucleoprotein complex (Fig. 3B).

Fig. 3.

Functional groups of genes significantly enriched among differentially expressed genes in rat lungs during either IH or SH. Functional groups of genes enriched among the clusters upregulated in IH (A), downregulated in IH (B), upregulated in SH (C), and downregulated in SH (D) are shown. Within the parentheses below each Gene Ontology (GO) functional group name, n_i, p_i%, cl. c represent the number (n_i) of the hypoxia-induced genes in GO group i, the percentage (p_i%) of the hypoxia-induced genes assigned to the GO group i (i.e., p_i% = n_i/N_i, where N_i is the total number of the genes in GO group i), and the cluster(s) (cl. c) significantly enriched with the genes in the same GO group i. GO groups in this figure satisfied 1) P value < 0.1 with the false discovery rate (FDR) control and 2) n_i > 10 or p_i > 20., Functional groups that satisfied P < 0.01 without the FDR correction but did not satisfy P < 0.1 with the FDR correction.

The majority of the enriched upregulated genes resided in the two clusters—clusters 1 and 2 (Fig. 3A). Specifically, genes in cluster 1 displayed a cyclically regulated pattern that peaked on days 1 (weakly) and 7 (strongly) (Fig. 2A and Supplemental Fig. E1A). Of all the clusters induced by IH, this cluster contained the largest number of significantly enriched functional groups of genes, which overlapped mostly with those enriched among all upregulated genes induced by IH. Interestingly, six of the eight genes encoding the steroid hormone receptors (including ESR1 and ESR2) induced by IH were coexpressed in this cluster and were significantly overrepresented (P = 0.02 with FDR).

Genes in cluster 2 exhibited a typical bell-shaped expression pattern with a single peak on day 7 (Fig. 2A and Supplemental Fig. E1A); this cluster was significantly enriched with those participating in K⁺ transport and neuronal activities. Since ion channels (especially K⁺ channels) are important for neuronal signaling, genes in this cluster were likely to mediate neuronal responses to IH in rat lungs.

ESR1-Regulated Genes Are Enriched Among Differentially Expressed Genes in IH

Since genes encoding steroid hormone receptors were overrepresented among the IH-induced genes, we postulated that ESR1 plays regulatory roles in hypoxic responses to IH. To test this possibility, we examined the differentially expressed genes induced by IH. To our surprise, we found 248 potential ESR1 target genes (Supplemental Table E3) (4); most of these resided in clusters 1, 2, and 6, which, interestingly, bore similar expression patterns (i.e., clusters 2 and 6 displayed almost the mirror image of one another). GO analysis revealed that the ESR1 target genes in the upregulated clusters primarily participated in the following activities: response to stress, ion transport, neurological system process, G protein-coupled receptor signaling, oxidoreductase activity, and steroid hormone receptor activity (Fig. 4A and Supplemental Fig. E3). Significantly, most of these functional groups overlapped with those found among the upregulated genes in IH, suggesting that ESR1 plays key regulatory roles during IH. The ESR1 target genes in the downregulated clusters were mainly involved in ubiquitin-dependent protein catabolic process, cytokine activity, and ribonucleoprotein complex (Fig. 4B).

Fig. 4.

Functional groups of genes significantly enriched among potential estrogen receptor 1 (ESR1) target genes in rat lungs during IH. The functional groups of genes (shown in black) enriched among potential ESR1 target genes in the upregulated clusters (A) and the downregulated clusters (B) (see Supplemental Fig. E1 for details) are shown. The functional groups overrepresented among all the upregulated genes (shown in Fig. 3A) but not among ESR1 target genes are shown in gray., Functional groups that satisfied P < 0.01 without the FDR correction but did not satisfy P < 0.1 with the FDR correction.

Verification of Genes Differentially Expressed Over Time in IH

To verify our findings from microarray analysis, we performed both literature survey and qRT-PCR. Many of the differentially expressed genes we identified in the present experiments have been reported previously to be inducible by hypoxia (9, 30, 42), which included the upregulated Nfkb1, Ho-2, and Htr5a and the downregulated Kcnma1. These genes were therefore not verified experimentally.

We validated differential expression of Esr1 and 11 other genes including 4 known ESR1 target genes: P2rx2, Areg, Bmp2, and Smad1 (4, 5, 23) as well as Kcnj6, Mapk1, Cacna1c, Nr0b1, Cyp26b1, Cyp2b15, and Atp1a2 with qRT-PCR. Consistent with our microarray results (Fig. 5A, left), we observed a cyclic pattern of Esr1 in rat lungs in response to IH, but not to SH, with qRT-PCR (Fig. 5A, right). The other 11 selected genes were also confirmed to be significantly differentially expressed in IH, and their expression patterns measured by qRT-PCR agreed well with those revealed by LRSA from microarray data (Fig. 5, B–L).

Fig. 5.

Verification of genes differentially expressed during IH. A, left: temporal expression pattern of the Esr1 gene (residing in cluster 1) fitted by local regression and spectral analysis (LRSA). Each black dot in the plot represents the expression level of the gene (shown on y-axis, relative to expression of Esr1 in RA control rats) in each individual rat exposed to IH for the designated days (shown on x-axis) in the normalized microarray data; solid curve shows the smoothing curve fitted to the microarray data by LRSA and dotted curves show the simultaneous 95% confidence intervals estimated for the fitted data. Right: expression level of Esr1 (relative to expression of Esr1 in RA control rats) measured by quantitative real-time PCR (qRT-PCR). Data for rats in response to RA or to IH or SH for 1, 3, 7, 14, or 30 days are shown. Values are means ± SE. Significant differences: *P < 0.05 compared with RA control rats. B–L: other verified IH-induced genes are Kcnj6, P2rx2, Mapk1, Areg, Cacna1c, Nr0b1, Cyp26b1, Cyp2b15, Atp1a2, Bmp2, and Smad1.

Regulatory Networks in IH

Using network analysis, we detected a regulatory network among differentially expressed genes in IH (Fig. 6A and Supplemental Fig. E4). The network contained ∼500 genes—the biggest network allowed to be identified by the IPA software. This network contained a few hub nodes, each of which formed direct connections with a large number of other nodes in the network (Fig. 6A and Supplemental Fig. E4A). It can be seen that ESR1 was one of the hub nodes in the network (shown in detail in Fig. 6B), which connected closely with other hub nodes including ARNT (also known as HIF-1β), VEGF (encoding vascular endothelial growth factor), NF-κB, JUN, MAPK1, as well as stress proteins including HSP70; it also downregulated the genes involved in transcriptional and splicing machinery such as Med20 and SF3A1. In addition, ESR1 interacted closely with the three other steroid hormone receptors that were also differentially expressed in IH [ESR2, androgen receptor (AR), and nuclear receptor NR0B1], suggesting that there was cross talk between these proteins in the network.

Fig. 6.

Gene regulatory network detected in rat lungs during IH. A: graphical representation of the molecular relationships between nodes (i.e., genes and/or gene products) in the network identified in rat lungs exposed to IH for 7 days. An edge (line) represents the biological relationship between 2 nodes. The intensity of the node color indicates the magnitude of the upregulation (red) or downregulation (green) of the gene at a specific time point relative to its expression at the control time point (day 0). The direct connections of ESR1 with other nodes are shown by blue lines. Additional views of this network can be seen in Supplemental Fig. E4. B: interaction of ESR1 with other nodes in the identified network shown in A.

Interestingly, genes in this network have been previously associated significantly with cancer (P = 1E-19), hematologic disorder (P = 6E-18), cardiovascular disorder (P = 3E-12), gastrointestinal disease (P = 4E-10), as well as inflammatory, skeletal and muscular, connective tissue, genetic, neurological, and behavior disorders (Supplemental Fig. E5A and Supplemental Table E4). Furthermore, our analysis showed that ESR1 has been directly associated with most of these disorders (Supplemental Fig. E5A and Supplemental Table E4).

Temporal Gene Expression Patterns in SH

We also analyzed temporal expression data obtained from rats exposed to SH and detected 936 differentially expressed genes induced by SH (Supplemental Table E5). These genes displayed six temporal expression patterns (Fig. 2C), and many of them were differentially expressed in an acute manner (Fig. 2C and Supplemental Fig. E1B). Examination of the genes induced over time in SH showed that a majority of the genes were induced on day 1, while only a few genes were induced at later time points (days 14 and 30) (Fig. 2D).

GO analysis revealed that the upregulated clusters (clusters 1–3, see Supplemental Fig. E1 for details) in SH were significantly enriched with the functional group of genes involved in immune response; the same group of genes was also coexpressed and enriched in cluster 2 (Fig. 3C), while the downregulated clusters (clusters 5 and 6, Supplemental Fig. E1) were overrepresented by genes involved in RNA binding and processing (Fig. 3D). Our results also showed that cluster 1 was significantly overrepresented by genes involved in organ (particularly blood vessel) morphogenesis; since pulmonary vascular remodeling is a known effect of SH in rats, genes in this cluster might play roles in this process.

Despite the finding that gene expression patterns in response to IH and SH were distinctly different in rat lungs, we found that a considerable number of the genes were differentially expressed in both types of hypoxia (Supplemental Fig. E7A). It appeared that the genes upregulated in both IH and SH were significantly overrepresented by those participating in system development, anatomic structure morphogenesis, and cell cycle (Supplemental Fig. E7B), while the downregulated genes were overrepresented by those involved in pyrophosphatase and ATPase activities (Supplemental Fig. E7C).

Verification of Genes Differentially Expressed in Rat Lungs During SH

Many differentially expressed genes we identified in SH have been previously reported to be inducible in hypoxia, which included Edn1 and Bmpr2, whose differential expression has been associated with pulmonary hypertension developed in animals exposed to SH (35, 37). It is unknown, however, how these genes were differentially expressed over time in response to hypoxia exposure.

We validated the temporal expression patterns of the eight genes with qRT-PCR. It can be seen that Edn1 exhibited a cyclic temporal pattern that peaked on days 1 (strongly) and 7 (weakly) in microarray data (Fig. 7A, left). We confirmed this expression pattern of Edn1 in rats exposed to SH with qRT-PCR (Fig. 7A, right). The significant downregulation of Bmpr2 over time was also confirmed in rats exposed to SH with qRT-PCR (Fig. 7G). However, unexpectedly, the downregulation of Bmpr2 was also observed in rats exposed to IH with qRT-PCR (Fig. 7G, right), despite the fact that we did not detect this in the microarray data. The expression patterns of the six other genes, Bmp2, Mmp9, Cd8b, Cyp26b1, Cacna1c, and Smad1, were also confirmed by qRT-PCR (Fig. 7, B–H).

Fig. 7.

Verification of genes differentially expressed during SH. A, left: temporal expression pattern of the Edn1 gene (residing in cluster 1) fitted by LRSA. Details of the plot can be found in Fig. 4 legend. Right: expression level of Edn1 (relative to its expression in RA control rats) measured by qRT-PCR. Data for rats in response to RA or to IH or SH for 1, 3, 7, 14, or 30 days are shown. B–H: verification of other genes differentially expressed in SH: Bmp2, Mmp9, Cd8b, Cyp26b1, Cacna1c, Bmpr2, and Smad1. Data are means ± SE. Significant differences: *P < 0.05 compared with RA control rats.

Regulatory Networks in SH

Our network analysis revealed a regulatory network activated by SH (Fig. 8 and Supplemental Fig. E8). This network contained the hub nodes including VEGF, NF-κB, and TGF-β, all of which have been implicated in vascular remodeling and pulmonary hypertension during hypoxia (39), as well as EDN1, IL1B, MMP9, and PI3K. Unlike the IH-induced network, ESR1 was not involved in this network.

Fig. 8.

Gene regulatory network detected in rat lungs during SH. Plot illustrates the molecular relationships between genes and/or gene products in the network identified in rat lungs in response to SH for 1 day. The details on the graphical representation of the network can be found in Fig. 6 legend. Additional views of this network can be seen in Supplemental Fig. E8.

Finally, disease relevance analysis revealed that this network is significantly associated with physiological malfunctions including cancer (P = 2E-18), hematologic disease (P = 3E-16), cardiovascular disorder (P = 2E-8), colorectal cancer (P = 2E-12), and skeletal and muscular disorder (P = 2E-7) (Supplemental Fig. E5B, Supplemental Table E6). These abnormalities were also associated with the IH-activated network; however, our analysis indicated that a majority (more than two-thirds) of the genes associated with each malfunction were unique to either IH or SH, suggesting that different molecular mechanisms were underlying the same diseases associated with these two different types of hypoxia (Supplemental Table E6). In addition, the SH-activated network is also significantly enriched with genes associated with rheumatic disease (particularly arthritis), autoimmune disease, and endocrine system disorders (particularly diabetes) (Supplemental Fig. E5B and Supplemental Table E6), which were specific to SH, but not to IH.

DISCUSSION

We employed a systems biology approach to study temporal effects of IH and SH by examining time-dependent gene expression profiles from whole lung tissues of rats exposed to either IH or SH. Our results demonstrate that IH and SH induced distinct sets of genes with different temporal expression patterns. IH induced genes mostly involved in ion transport and homeostasis, neurological processes, system development, and steroid hormone receptor activity, while SH induced genes principally participating in immune responses. The IH-activated network suggested a role for cross talk between ESR1 and other key proteins in hypoxic responses (e.g., ARNT and MAPK1) during IH, whereas the SH-activated network was indicative of vascular remodeling and pulmonary hypertension. We confirmed the temporal changes in the expression of 12 genes (including Esr1 and 4 other ESR1 target genes) in IH and 8 genes in SH with qRT-PCR. Our data suggest that ESR1 may be a regulator of lung hypoxic response to IH.

Several lines of evidence have emerged from our analyses suggesting that ESR1 is a potential key regulator that modulates the expression of the genes responsible for major hypoxic responses to IH. First, we verified the differential expression of the four ESR1 target genes during IH, P2rx2, Bmp2, Areg, and Smad1, with qRT-PCR. These genes participate in diverse activities in various tissues where ESR1 plays important roles: P2X2 is involved in neuronal responses to hypoxia in carotid body; BMP2 is implicated in bone development (6); amphiregulin is a key mediator of ESR1 function in mammary gland development (5); and the direct contact of Smad1 with ESR1 is implicated in the tumorigenesis of pituitary prolactinoma (23). Additionally, we identified 248 ESR1 target genes among genes induced by IH, and they were enriched with the same functional groups enriched in all upregulated genes during IH (Fig. 4A and Supplemental Fig. E3). Significantly, most of these activities have been well established to mediate important downstream effects of ESR1 signaling and action (12, 14, 17, 38). Furthermore, our network analysis illustrated that ESR1 interacts closely with numerous genes in the IH-activated network (Fig. 6B). Finally, ESR1 interacts directly with AKT, MAPK1, and PI3K in the IH-activated network; these protein kinases are known to modulate the activity of ESR1, and thus allow synergistic regulation of gene expression by these signaling pathways (18). All of the above-described evidence suggests that ESR1 plays major regulatory roles in rat lung in response to IH. This finding, however, is both interesting and unexpected, since all the rats used in this work were male. Indeed, there is a recent awareness that ESRs play important roles in biological processes in male animals, including vascular homeostasis and protection (17) and bone metabolism (34); there has been no report that they play key regulatory roles in the context of diverse hypoxic adaptive or injury-relayed processes in male animals.

Our analyses also suggest several additional important findings. First, in the detected network induced by IH, ESR1 interacted closely with other hub nodes previously known to play key roles in hypoxia and/or ESR signaling, such as ARNT, VEGF, MAPK1, and EDN1, suggesting that there is interplay between ESR1 and these important proteins during lung hypoxic responses to IH. Additionally, our disease relevance analysis showed that ESR1 is directly involved in many disorders including cancer, hematologic, cardiovascular, and inflammatory disorders, and neurological, behavior, and metabolic disorders (Supplemental Fig. E5A). Most notably, these are also the disorders associated with patients with sleep-disordered breathing (33). These results are both interesting and potentially important—they suggest that ESR1 may not only mediate crucial hypoxic responses to IH in the animal but also be directly involved in pathophysiological mechanisms underlying abnormalities associated with patients with sleep-disordered breathing.

In this context it is worth noting the potential effect of ESR1 on regulating neuronal responses in rat lungs during IH. Our data showed that genes participating in neuronal processes were overrepresented among the potential target genes of ESR1 in hypoxia. Some of these genes are known to mediate crucial hypoxic responses in the brain and the carotid body. For example, increasing evidence has shown that the purinergic receptor P2X2 is one of the earliest proteins responding to hypoxia in mammals for mediating adaptive changes in breathing in the carotid body. We have identified and confirmed upregulation of the P2rx2 gene (encoding P2X2) in rats exposed to IH in microarray data and with qRT-PCR and thus provide initial evidence that the expression of P2rx2 is induced in the rat lung during IH. Besides, we have identified upregulation of the genes encoding the glutamate receptor (Grik5) and the dopamine receptor D4 (Drd4)—both of which are receptors for excitatory neurotransmitters and critical proteins involved also in adaptive changes in breathing in the central nervous system (CNS) (8, 22). In addition, genes encoding receptors for important neurotransmitters known to be active in the brain have also been identified as upregulated in IH and potentially regulated by ESR1, which included Gabrr2 (encoding an inhibitory GABA receptor A), Chrnb2 (encoding a nicotinic cholinergic receptor), and Adra1d (encoding an adrenergic α-receptor). In particular, upregulation of the nicotinic cholinergic and adrenergic receptors has been associated with increased heartbeat and systemic hypertension: both symptoms have been reported in patients with sleep-disordered breathing owing to IH (33). These results are also consistent with previous studies indicating that ESR1 mediates neuroprotection by modulating expression of neuronal genes (18) but so far have not been demonstrated in the lung.

An interesting feature revealed by our GO analysis of IH gene expression patterns is that while a large number of functional groups were significantly overrepresented in upregulated clusters, only a few were overrepresented in downregulated clusters. This is consistent with previous observations suggesting that when animals are exposed to hypoxia specific responses essential to survival are elicited, while nonessential biological processes are shut off (1). Our data suggest that this is indeed the case. First, GO analysis showed that the overrepresented groups in the upregulated genes are underrepresented in the downregulated genes, suggesting that essential functional groups are indeed mostly upregulated during IH. Second, there were significantly fewer (and no additional) functional groups enriched among all differentially expressed genes during IH compared with those enriched in the upregulated clusters, which demonstrates again the diluting effect of the downregulated clusters (i.e., the absence of the essential functional groups in these clusters) and also indicates that the up- and downregulated clusters contain genes involved in different biological processes.

The comparison of SH to IH yielded some meaningful differences. Compared with IH, our results showed that 1) SH induced a different set of genes in rat lungs, most of which displayed acute patterns that peaked on day 1—this may just reflect the dose effect of hypoxia during IH and SH, that is, the more severe exposure of hypoxia during SH may lead to more acute responses in animals; 2) the SH-induced genes were overrepresented by those involved in immune response; 3) there was no obvious involvement of ESR1 in the SH-activated network; and 4) disease relevance analysis revealed that the SH-activated network is specifically, significantly associated with rheumatic disease and endocrine system disorders (particularly diabetes). Despite the finding that IH and SH elicited distinct hypoxic responses in rat lungs, there were a considerable number of common genes induced by both types of hypoxia. Most of the common genes were downregulated in IH and SH and were enriched with the functional groups involved in pyrophosphatase and ATPase activities, suggesting that reduction of energy consumption is a common hypoxic response in both types of hypoxia.

An interesting attribute of the gene expression patterns observed in our study is that many genes induced in IH or SH seem to display homeostatic patterns, that is, the expression of the genes mostly returned to their baseline values after induction at the early time points. These results are in contrast to our physiological data, which showed that systemic BPs were increased continuously in rats exposed to IH and that RVPs were also elevated in rats during either IH or SH, albeit at different levels (Fig. 1). One possible explanation for this discrepancy is that the genes required for inducing the increase in systemic BP and RVP in rat lungs are different from those required for maintaining them. Indeed, we can see that in the IH- or SH-induced networks (Supplemental Figs. E4A and E8A) genes differentially expressed during hypoxia (i.e., nodes shown in red or green) can not only affect expression of each other but also modulate expression of proteins whose mRNAs were not differentially expressed (i.e., nodes shown in white) during early hypoxic response. These proteins may in turn affect the physiology of the animals long after the expression of the majority of genes goes back to the original level. A similar phenomenon has been observed in other physiological responses during hypoxia. For example, the activation of a serotonin receptor (5-HT_2A) is required for the initiation but not the maintenance of long-term facilitation elicited during IH, yet other proteins are required for maintaining long-term facilitation in the same animals (2, 8, 19). While our study is unique in that we studied temporal gene expression patterns in response to IH or SH up to 30 days, our results also suggest that to fully understand the long-term complex regulation of the response to IH or SH, temporal gene expression patterns beyond 30 days should be studied.

In this study we aimed to provide a systems-level view of the lung response to IH and SH. We identified global expression patterns that were distinct to these types of stimuli and validated some of the genes that represented these patterns and our key biological insights. It is important to note that the lung is a complex and highly dynamic organ, and thus it would be important to determine the cellular populations that express these changes in gene expression. While such an extensive localization effort was beyond the scope of this work, we believe that the wide availability of our results will encourage scientists and investigators to identify and explore these cellular response niches.

We have attempted here to identify molecular mechanisms underlying hypoxic responses in the rat lung during IH or SH with a systems biology approach. While previous studies on hypoxia have mainly focused on the hypoxic effects on the CNS and cardiovascular system, our work provides the first comprehensive genomewide investigation of the hypoxic effects of IH and SH on the rat lung; it demonstrates the wealth of the global transcriptional response of the lung to IH and SH and suggests that the lung may participate in the downstream physiological effects of IH and SH. Our findings also suggest that ESR1 is a regulator of hypoxic responses in IH but not in SH and that genes critical for hypoxic responses in the brain and cardiovascular system may also play important and active roles in the lung. Our results provide clues and insights into novel molecular mechanisms underlying IH and SH responses, thereby warranting future exploration of such novel pathways.

GRANTS

W. Wu's work was funded by National Institutes of Health (NIH) Grant P50-HL-084932 and National Science Foundation Grant CCF-0523757. N. Kaminski's work was funded by NIH Grants HL-073745, HL-0793941, and HL-0894932. N. Kaminski was also supported by a generous donation from the Simmons Family. N. B. Dave's work was funded by NIH Grant 1-F32-HL-78164-2. D. Gozal is supported by NIH Grants HL-65270, HL-69932, and SCOR 2P50-HL-60296-06, The Children's Foundation Endowment for Sleep Research, and the Commonwealth of Kentucky Challenge for Excellence Trust Fund.

DISCLOSURES

D. Gozal serves on the national speaker bureau for Merck Company and has received honoraria for lectures in 2006 and 2007.