Lung omics signatures in a bronchopulmonary dysplasia and pulmonary hypertension-like murine model

Abstract

Bronchopulmonary dysplasia (BPD), the most common chronic lung disease in infants, is associated with long-term morbidities, including pulmonary hypertension (PH). Importantly, hyperoxia causes BPD and PH; however, the underlying mechanisms remain unclear. Herein, we performed high-throughput transcriptomic and proteomic studies using a clinically relevant murine model of BPD with PH. Neonatal wild-type C57BL6J mice were exposed to 21% oxygen (normoxia) or 70% oxygen (hyperoxia) during postnatal days (PNDs) 1–7. Lung tissues were collected for proteomic and genomic analyses on PND 7, and selected genes and proteins were validated by real-time quantitative PCR and immunoblotting analysis, respectively. Hyperoxia exposure dysregulated the expression of 344 genes and 21 proteins. Interestingly, hyperoxia downregulated genes involved in neuronal development and maturation in lung tissues. Gene set enrichment and gene ontology analyses identified apoptosis, oxidoreductase activity, plasma membrane integrity, organ development, angiogenesis, cell proliferation, and mitophagy as the predominant processes affected by hyperoxia. Furthermore, selected deregulated proteins strongly correlated with the expression of specific genes. Collectively, our results identified several potential therapeutic targets for hyperoxia-mediated BPD and PH in infants.

INTRODUCTION

Bronchopulmonary dysplasia (BPD) is the most common chronic lung disease in preterm infants resulting from interrupted lung development (21). Pulmonary hypertension (PH) develops in 17% of patients with BPD and increases BPD-associated mortality and morbidity (2). However, this complex disease lacks specific therapies, and the molecular mechanisms underlying BPD and PH remain unclear.

Supplemental oxygen and mechanical ventilation are life-sustaining therapies administered to infants with respiratory failure. However, oxidative stress and inflammation associated with these therapies disrupt normal lung development and lead to BPD and PH in both infants and mice (21, 30, 36). Murine lungs at birth are equivalent to the developmental lung stage of a 25- to 26-wk-old infant (42). Therefore, mice are the most commonly used laboratory animal to model human BPD. Several preclinical studies, including ours, have shown that exposure of neonatal mice to prolonged hyperoxia leads to a phenotype that models BPD and PH in infants (3, 19, 35).

Gene expression microarray and reverse-phase protein array (RPPA) allow quantification of the transcriptome and proteome, respectively. These powerful tools can identify novel genes, proteins, biological pathways, and their interactions and can help us understand the molecular mechanisms underlying disease processes. Gene-based assays such as RNA microarrays and RNA sequencing have been commonly utilized to identify the molecular mechanisms of BPD (1, 5–8, 12, 17, 26, 31, 40); however, proteomic assays are rarely used (15, 25, 38) to unravel BPD pathogenesis. Proteomic studies are important for several reasons. First, proteins are the final mediators of biological processes; therefore, they represent meaningful therapeutic targets. Second, mRNA levels may not always correlate with the level and function of corresponding proteins because of processes such as posttranscriptional modifications. Finally, combined transcriptomic and proteomic analysis can further elucidate the molecular mechanisms underlying disease. Hence, the objective of this study was to determine and correlate gene and protein expression profiles in a clinically relevant mouse model of hyperoxia-induced lung injury, to identify suitable therapeutic targets for infants with BPD with PH. Using neonatal C57BL6J wild-type (WT) mice, we tested the hypothesis that hyperoxia exposure during the saccular and alveolar development phase alters the level of genes and proteins necessary for optimal lung development and repair.

MATERIALS AND METHODS

Animals, hyperoxia experiments, and tissue collection.

This study was approved by the Institutional Animal Care and Use Committee of Baylor College of Medicine and conducted as per American Physiological Society (2010–2011) guidelines for animal studies. C57BL6J WT male and female pups from multiple litters were pooled before being randomly and equally redistributed to the dams before being exposed to 21% oxygen (normoxia; controls) or 70% oxygen (hyperoxia) during postnatal days (PNDs) 1–7. There were 10 pups in each experimental group (array experiments: n = 4/group; nonarray experiments: n = 6/group). Hyperoxia experiments were conducted in plexiglass chambers into which oxygen was delivered continuously through an oxygen blender to achieve a constant level of 70% oxygen. The dams were rotated between air- and hyperoxia-exposed litters every 24 h during the hyperoxia exposure period to prevent oxygen toxicity in the dams and to control for maternal effects between the groups (35). At PND 7, the animals were euthanized with intraperitoneal injections of 200 mg/kg of sodium pentobarbital, and the lung tissues were collected for genomic and proteomic studies.

Microarray analysis.

RNA from mouse lung tissue exposed to normoxia or hyperoxia (n = 4 per group) was purified using the Direct-zol RNA MiniPrep kit (Zymo Research, CA). Microarray experiments were performed using the Illumina Sentrix Beadchip Array Mouse Ref8_V2. The BeadStudio analysis software (Illumina, CA) was used to apply the standard quality control thresholds and analyze the data.

Differential gene expression analysis.

The effect of oxygen exposure on gene expression was assessed using a t-test. Differentially expressed genes (DEGs) were considered statistically significant at a false discovery rate (FDR)-adjusted P value < 0.01 and fold change (FC) exceeding 1.25×. Gene set enrichment analysis (GSEA) was performed using GSEA version 3. Enrichment scores ≥0.4 (positive or negative) were considered significant. Furthermore, Enrichr tool was also used for functional classification of the DEGs as described previously (10, 23). The data set will be deposited in the National Center for Biotechnology Information’s Gene Expression OMNIBUS.

Real-time quantitative PCR assays.

To validate and reproduce microarray data, we performed real-time quantitative PCR (RT-qPCR) using lung RNA samples from the microarray experiment and a separate set of samples from mice exposed to normoxia or hyperoxia (n = 6 per group). RT-qPCR analysis was performed using the TaqMan gene expression master mix and gene-specific primers (Grand Island, NY).

RPPA analysis.

Protein lysates from mice exposed to normoxia or hyperoxia (n = 3 per group) were prepared using the modified Tissue Protein Extraction Reagent (Pierce) (20). RPPA analysis was then conducted as previously described (20). Briefly, lysates were probed with a set of 204 antibodies. Antibody slides that failed quality inspection were either retested or removed. The remaining 182 antibodies (116 to detect total protein and 66 to detect phosphorylated states of proteins) were used for subsequent analyses.

Differential protein expression analysis.

The heatmap of median-centered, normalized expression values and unsupervised hierarchical clustering values was generated using the R statistical software. Significant differences in protein expression levels between groups were determined using a t-test (P < 0.05 and at least 1.2 × FC). Enrichment analysis of differentially expressed proteins (DEPs) was performed using gene ontology (GO) mapping. The global mouse gene database was used as a background for the GO enrichment analysis, and a gene set was considered significantly enriched at an FDR-adjusted P value < 0.05.

Immunoblot assays.

To validate and reproduce RPPA data, we performed immunoblot assays using protein lysates from the RPPA experiments and from a separate set of samples from mice exposed to normoxia or hyperoxia (n = 6 per group). Immunoblotting analysis was performed using primary antibodies against β-actin (Santa Cruz Biotechnology), phosphorylated (p)-AMP-activated protein kinases (p-AMPK) a[T172], p-AMPKb1[S108], SLUG, and platelet-derived growth factor receptor (PDGFR) b (Cell Signaling Technology).

Correlation between proteomic and transcriptomic data.

Correlation analysis between the relative abundance of three DEPs (p-AMPKb1[S108], PDGFRb, and SLUG), and whole-genome transcriptional profiles in normoxic and hyperoxic lungs were analyzed using RPPA and microarray data. The most significantly correlated transcripts were identified using correlation coefficient cutoffs > ± 0.95 and P < 0.05.

Statistical analyses.

The results were analyzed using the GraphPad Prism 5 software. Data are expressed as means ± standard deviations (SD). An unpaired t-test was used to determine differences in mRNA and protein levels between normoxia- and hyperoxia-exposed mice. P < 0.05 was considered significant.

RESULTS AND DISCUSSION

In the present study, we analyzed the effects of hyperoxia on lung genomic and proteomic profiles in a murine model of BPD and PH using high-throughput technologies and found significant correlations between transcriptomic and proteomic data.

WT neonatal mice exposed to 70% oxygen show persistent structural lung abnormalities up to PND 56 (28, 29, 35), PH at PND 14 (35), and PH and right ventricular dysfunction at PND 56 (28). Using this disease model, we conducted transcriptomic analysis on PND 7. Hyperoxia-mediated molecular and biochemical perturbations that lead to BPD phenotype are early events in the disease pathogenesis. Moreover, identifying these perturbations early in the disease process will help us discover biomarkers that have the potential to prevent and/or treat BPD. Therefore, we analyzed the lung tissues at PND 7 rather than at PND 14 to increase the likelihood of identifying the differentially regulated genes, proteins, and pathologic pathways. Analysis of DEGs between controls and hyperoxia-exposed mice showed 344 genes significantly regulated by hyperoxia at an FDR-adjusted P < 0.01 and FC ≥ 1.25 (Supplemental Table S1; Supplemental Material for this article is available online at the Journal website). In hyperoxia-exposed mice, 125 genes were downregulated and 219 genes were upregulated, compared with those in controls. GSEA revealed that hyperoxia negatively enriched the cellular compartmental process integral to the plasma membrane (Fig. 1A) and positively enriched biological and molecular functional processes such as apoptosis and oxidoreductase activity, respectively (Fig. 1, B and C). Additional functional classification of the DEGs using the Enrichr tool is shown in Supplemental Table S2. Based on their relevance in oxygen toxicity and lung development, we selected six genes [ATP binding cassette subfamily C member 3 (ABCC3), BCL2 binding component 3 (BBC3), cyclin dependent kinase 5 regulatory subunit 1 (CDK5R1), glutathione peroxidase 2 (GPX2), neurotrophic receptor tyrosine kinase 2 (NTRK2), and platelet derived growth factor receptor alpha (PDGFRA)] for microarray data validation (using microarray samples) and reproducibility (using nonarray samples; additional biological replicates) by RT-qPCR. Microarray data were validated and reproduced for five of six (83.3%) genes tested (Fig. 1D). The minor discrepancies between microarray and RT-qPCR analysis may be related to their different sensitivities and gene expression detection mechanisms. The magnitude change in expression identified by microarray correlated with that of the RT-qPCR data. Consistent with other studies (4, 6, 9, 24, 27, 33), hyperoxia dysregulated genes that modulate cell migration, proliferation and death (ASAH3L, BAX, CDKN1, COL18A1, CXCL17, GDF15, MEST, MT1, NUPR1, PDGFR, SLC19A2, TRIB3), and oxidative stress response (GPX2, GPX3, GSTA3, GSTA4, GSTM1, GSTM2, GSTM3, MGST1), which are biological processes that contribute to BPD pathogenesis. Identifying new genes that regulate these processes may enable the development of a comprehensive gene profile of biomarkers or therapeutic drug targets for infants with BPD. Therefore, we chose to validate some of these genes that are known and unknown to be dysregulated by hyperoxia. Consistent with similar studies in rodents (6, 14, 27) and infants with BPD (34), we observed that hyperoxia increased the expression of genes that regulate oxidative stress (GPX2), although decreasing the expression of those that facilitate cell proliferation and organogenesis (NTRK2 and PDGFA). Additionally, we found genes not previously shown to be dysregulated in experimental or human BPD. BCL2 binding component 3 (BBC3) expression was upregulated by hyperoxia. As a member of the proapoptotic subclass of BCL2 protein family, uninhibited BBC3 expression induces mitochondrial dysfunction and activation of apoptosis signaling cascades. Inhibition of BBC3 is shown to protect human intestinal cells against hyperoxic injury (39). On the other hand, CDK5R1, which is known to regulate neuronal maturation (13) was downregulated by hyperoxia. The decreased expression of genes that facilitate neuronal development and maturation (NTRK2 and CDK5R1) under hyperoxia is noteworthy. NTRK2 is expressed in several lung cell types and is shown to be necessary for normal alveolar and nervous system development of murine lungs (16). Furthermore, CDK5R1 is also shown to affect lung cell proliferation under hypoxic conditions (22). Therefore, decreased expression of these genes may represent some of the mechanisms through which hyperoxia contributes to interrupted development and abnormal innervation and contractile response of the lungs. It is also possible that hyperoxia could also dysregulate these genes in the brain, which would partly explain the mechanism underlying BPD-associated neurological morbidities. Therefore, optimizing the level and function of BBC3, NTRK2, and CDK5R1 in the lung cells has the potential to preserve mitochondrial function, mitigate apoptosis and abnormal airway and pulmonary vascular contractility, and facilitate optimal lung development and repair under hyperoxic conditions.

Fig. 1.

Lung transcriptomics of 1-wk-old mice exposed to hyperoxia. Lung tissues of mice exposed to normoxia or hyperoxia for 7 days were analyzed using microarrays. GSEA of microarray data showing significant negatively (A) and positively (B and C) correlated cellular compartmental and biological and molecular function processes, respectively, in the hyperoxia group (FDR < 0.01; P < 0.05). RT-qPCR quantification of ABCC3, BBC3, CDK5R1, GPX2, NTRK2, and PDGFRA mRNA levels in samples from microarray experiments (n = 3 per group) and a separate set of samples (nonarray samples; n = 6 per group) (D). Values are presented as means ± SD *Significant differences between normoxia and hyperoxia groups at P < 0.05 (t-test). FDR, false discovery rate; GSEA, gene set enrichment analysis; RT-qPCR, real-time quantitative PCR; see main text for relevant definitions of genes.

The lung proteomic profile of hyperoxia-exposed mice was assessed using RPPA on PND 7. A heatmap of the median-centered, normalized expression levels of DEPs demonstrated that lung protein expression profiles of control mice were distinguishable from those of hyperoxia-exposed mice (Fig. 2A). Analysis of DEPs between the two groups showed that 63 out of 182 proteins were differentially regulated by hyperoxia (Fig. 2A); however, only 21 out of 182 proteins were significantly regulated (P < 0.05) by hyperoxia with FC exceeding 1.2× (Supplemental Table S3). Among DEPs, 6 proteins were downregulated (Fig. 2B) and 15 were upregulated (Fig. 2C) by hyperoxia. The biological processes affected by hyperoxia were determined by performing enrichment analysis using the GO database. Downregulated proteins were enriched for organ growth, vasculogenesis, kidney development, male genitalia development, vascular smooth muscle differentiation, and vascular morphogenesis, indicating their repression by hyperoxia (Fig. 2D). In contrast, upregulated proteins were significantly enriched for processes specific to cytokine signaling, oxidative stress response, neuron apoptotic process, negative regulation of mitochondrial organization, and epithelial cell proliferation and migration, indicating activation of these pathways by hyperoxia (Fig. 2E). Based on their relevance in organ development and hyperoxic injury, we selected four DEPs (p-AMPKa[T172], p-AMPKb1[S108], SLUG, and PDGFRb) for validation (using microarray samples) and reproducibility (using nonarray samples; additional biological replicates) of the RPPA data using immunoblotting analyses. RPPA data were validated and reproduced for all tested proteins (Fig. 3, A–F), and magnitude change in protein expression was similar in RPPA and immunoblotting experiments. Hyperoxia exposure did not affect AMPKα (23A3) protein levels (Fig. 3, A and D), suggesting that hyperoxia primarily increases the phosphorylation of and not the endogenous levels of AMPKα protein. Our findings reemphasize that insults such as hyperoxia during lung development alters distinct cytoprotective, inflammatory, developmental, and reparative lung responses (6, 9, 27, 33). We observed significant activation of AMPK under hyperoxia. Recently, an AMPK agonist, metformin, was shown to decrease hyperoxic lung injury in neonatal rats (11). Although the main function of this evolutionarily conserved kinase is to maintain energy homeostasis (18), it also contributes to other cellular processes, including growth, autophagy, mitophagy, and redox regulation. Autophagy and mitophagy have been recently identified as major contributors to BPD pathogenesis (37, 41). Furthermore, endothelial AMPK expression is decreased in adults with PH, and its function within the endothelium is necessary and sufficient to protect against hypoxia-induced PH in adult mice (32). Therefore, strategies directed to increase AMPK activation has the potential to promote lung vascular health and mitigate hyperoxia-induced BPD and PH. In addition, we found that hyperoxia decreases the level of proteins critically involved in normal lung development (PDGFRb and SLUG). Although PDGFR expression was previously shown to be decreased in experimental and human BPD (34), the present study identified other novel proteins dysregulated by hyperoxia, such as SLUG.

Fig. 2.

Lung proteomics of 1-wk-old mice exposed to hyperoxia. Lung tissues of mice exposed to normoxia or hyperoxia (n = 3/group) for 7 days were analyzed using RPPA. Heatmap of unsupervised hierarchical clustering of hyperoxia-exposed mice inclusive of significantly DEPs (P < 0.05) (A). Significantly downregulated proteins (B) and upregulated proteins (C) by hyperoxia. GO analysis of significantly downregulated (D) and upregulated (E) biological processes by hyperoxia (FDR < 0.05; P < 0.05). FDR, false discovery rate; DEP, differentially expressed protein; GO, gene ontology; RPPA, reverse-phase protein array; see main text for relevant definitions of genes.

Fig. 3.

Validation of the RPPA results in immunoblot assays: Lung tissues of mice exposed to normoxia or hyperoxia for seven days from RPPA experiments (array samples; n = 3 per group) and a separate set of experiments (nonarray samples; n = 6 per group) were analyzed in immunoblot assays to determine the protein expression levels of p-AMPKα [Thr172], p-AMPKβ1 [Ser108], AMPKα [23A3], PDGFRβ, and SLUG (A). Densitometric analyses wherein p-AMPKα [Thr172] (B), p-AMPKβ1 [Ser108] (C), AMPKα [23A3] (D), PDGFRβ (E), and SLUG (F) band intensities were quantified and normalized to β-actin. The values shown are presented as the mean ± SD *Significant differences between normoxia and hyperoxia groups at P < 0.05 (t-test). AMPK, AMP-activated protein kinase; RPPA, reverse-phase protein array; see main text for relevant definitions of genes.

Proteins significantly dysregulated by hyperoxia are expected to be functionally associated with expression levels of genes that mediate responses to differential oxygen tension in the lungs. Identifying and understanding these functional relationships could shed light on the molecular basis of hyperoxia-mediated lung injury and enable early diagnosis and improved management. As a preliminary analysis, we selected three DEPs (p-AMPKb1[S108], PDGFRb, and SLUG) based on their relevance in developmental lung injury and performed a cross-platform transcorrelation analysis (protein/transcript) to identify genes whose expression most significantly correlated with the levels of these DEPs. This analysis was performed using RPPA and gene expression analytes from normoxic and hyperoxic lungs. We observed strong correlations (r > ± 0.95, P < 0.05) between selected DEP levels and the expression of specific genes (Tables 1 and 2). Determining whether these correlations indicate functional relationships between DEPs and gene expression is one of our future goals. The canonical functions of many of these genes suggest they may play a crucial role in the pathogenesis of hyperoxic lung injury.

Table 1.

Positively correlating gene transcripts. Gene transcripts showing a positive correlation with DEP relative abundance in normoxic and hyperoxic lungs

Gene	Correlation Coefficient	P Value
p-AMPKb1 (S108) differentially expressed
Hyperoxia
ATP1B1	0.952	0.048
GSTA3	0.953	0.047
ALDH4A1	0.962	0.038
BC048546	0.964	0.036
AHNAK	0.964	0.036
CLEC2D	0.964	0.036
ALDH1A1	0.968	0.032
AK1	0.970	0.030
CCNG1	0.971	0.029
CDKN1A	0.972	0.028
IFIT3	0.973	0.027
FAS	0.977	0.023
LYVE1	0.982	0.018
MDM2	0.985	0.015
CCND1	0.987	0.013
GAA	0.991	0.009
LMNA	0.992	0.008
HSPA9	0.994	0.006
EGR1	0.995	0.005
Normoxia
HOXB5	0.960	0.040
LAMP2	0.966	0.034
KISS1	0.967	0.033
DCXR	0.974	0.026
MEST	0.978	0.022
CDCA7	0.985	0.015
GSTM2	0.994	0.006
GYPA	0.996	0.004
COX6B2	0.998	0.002
PDGFRb (28E1) differentially expressed
Hyperoxia
FAHD2A	0.987	0.013
Normoxia
MEGF10	0.954	0.046
CCNG1	0.957	0.043
GPX3	0.960	0.040
CYP2F2	0.963	0.037
GSTA3	0.969	0.031
ADCY8	0.976	0.024
AOX3	0.980	0.020
GLB1L2	0.981	0.019
HLX	0.984	0.016
MGP	0.991	0.009
ATP1B1	0.998	0.002
ABCC3	0.999	0.001
Slug (C19G7) differentially expressed
Hyperoxia
ASAH3L	0.978	0.022
MGLL	0.989	0.011
Normoxia
GYPA	0.959	0.041
GSTM2	0.963	0.037
MFAP2	0.966	0.034
GJA4	0.969	0.031
ASZ1	0.975	0.025
MGST1	0.980	0.020
MEST	0.987	0.013
CCND1	0.990	0.010
DCXR	0.997	0.003
CDS1	0.998	0.002

See main text for relevant definitions.

Table 2.

Negatively correlating gene transcripts. Gene transcripts showing a negative correlation with DEP relative abundance in normoxic and hyperoxic lungs

Gene	Correlation Coefficient	P value
p-AMPKb1 (S108) differentially expressed
Hyperoxia
HTRA1	−0.999	0.001
AGT	−0.996	0.004
HIST2H2AC	−0.994	0.006
HOXB5	−0.990	0.010
CD82	−0.989	0.011
DNAJC15	−0.985	0.015
EIF5	−0.979	0.021
HIST1H2AO	−0.975	0.025
HDC	−0.968	0.032
D8ERTD82E	−0.964	0.036
ADCY8	−0.951	0.049
Normoxia
MEGF10	−0.996	0.004
ATP6AP2	−0.991	0.009
ENPP5	−0.985	0.015
GSTA3	−0.974	0.026
GCNT2	−0.963	0.037
GUCY1B3	−0.961	0.039
PDGFRb (28E1) differentially expressed
Hyperoxia
BC034076	−0.983	0.017
KCNK3	−0.975	0.025
DNAIC1	−0.974	0.026
Normoxia
MEST	−0.986	0.014
DLK1	−0.983	0.017
EXOC4	−0.968	0.032
EIF5	−0.965	0.035
LAMP2	−0.964	0.036
HOXB5	−0.955	0.045
Slug (C19G7) differentially expressed
Hyperoxia
ADAM28	−0.996	0.004
MEGF10	−0.987	0.013
CCDC85B	−0.985	0.015
CLIC6	−0.984	0.016
LOC100046120	−0.979	0.021
CKMT1	−0.970	0.030
Normoxia
LMNA	−1.000	0.000
ANGPT2	−0.999	0.001
ENPP5	−0.979	0.021
GUCY1B3	−0.964	0.036
HSPB8	−0.961	0.039
AHNAK	−0.961	0.039
IGFBP2	−0.956	0.044

See main text for relevant definitions.

Our study differs from others because we profiled both transcriptomic and proteomic data and performed a transcorrelation analysis between them. However, our study has a few limitations, which we plan to address in the future. First, the mouse lungs at birth are functionally different than those of a preterm infant and a murine model of hyperoxic lung injury may not accurately model human BPD. Second, omics analysis at single time-points may not identify all dysregulated genes, proteins, and pathways. Additionally, the number of quantifiable protein by RPPA in our study was rather limited to those validated by our core laboratory and is biased toward those selected for the array. Unbiased methods such as global proteomics and phosphoproteomics are required to identify all the significant pathways that are altered upon exposure to hyperoxia. Third, we did not map dysregulated genes and proteins to specific lung cell types. Finally, sex-specific effects were not evaluated.

In summary, we identified several genes and proteins that are potential biomarkers and therapeutic targets for BPD infants with PH. Additionally, our transcriptomic data may partly explain the mechanisms underlying BPD-associated neurologic morbidities. Importantly, our data reiterate that BPD with PH is a complex disease mediated by multiple pathways and may require a comprehensive approach for disease prevention and mitigation.

GRANTS

This work was supported by National Institutes of Health Grants HD-073323 (to B. Shivanna) and P30-CA-125123 (to S. Huang), Cancer Prevention & Research Institute of Texas Proteomics & Metabolomics Core Facility Support Award RP170005 (to S. Huang), American Heart Association Award BGIA-20190008 (to B. Shivanna), and American Lung Association Award RG-349917 (to B. Shivanna).

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

A.K.S., V.Y.N.G., R.T.M., J.L.H., S.H., and B.S. conceived and designed research; A.K.S., R.T.M., and B.S. performed experiments; A.K.S., V.Y.N.G., R.T.M., J.L.H., S.H., and B.S. analyzed data; A.K.S., V.Y.N.G., R.T.M., J.L.H., S.H., and B.S. interpreted results of experiments; A.K.S., V.Y.N.G., R.T.M., J.L.H., S.H., and B.S. prepared figures; A.K.S., V.Y.N.G., R.T.M., J.L.H., S.H., and B.S. drafted manuscript; A.K.S., V.Y.N.G., R.T.M., J.L.H., S.H., and B.S. edited and revised manuscript; A.K.S., V.Y.N.G., R.T.M., J.L.H., S.H., and B.S. approved final version of manuscript.