Chronic inflammation is a molecular element of the metabolic syndrome and type 2 diabetes. Saturated fatty acids (SFAs) are considered to be an important proinflammatory factor. However, it is still incompletely understood how SFAs induce proinflammatory cytokine expression. Hereby we report that activating transcription factor (ATF) 4, a transcription factor that is induced downstream of metabolic stresses including endoplasmic reticulum (ER) stress, plays critical roles in SFA-induced interleukin-6 (Il6) expression. DNA microarray analysis using primary macrophages revealed that the ATF4 pathway is activated by SFAs. Haploinsufficiency and short hairpin RNA–based knockdown of ATF4 in macrophages markedly inhibited SFA- and metabolic stress–induced Il6 expression. Conversely, pharmacological activation of the ATF4 pathway and overexpression of ATF4 resulted in enhanced Il6 expression. Moreover, ATF4 acts in synergy with the Toll-like receptor-4 signaling pathway, which is known to be activated by SFAs. At a molecular level, we found that ATF4 exerts its proinflammatory effects through at least two different mechanisms: ATF4 is involved in SFA-induced nuclear factor-κB activation; and ATF4 directly activates the Il6 promoter. These findings provide evidence suggesting that ATF4 links metabolic stress and Il6 expression in macrophages.
Chronic inflammation is a molecular element of the metabolic syndrome and type 2 diabetes. Several proinflammatory signaling pathways, including interleukin 6 (IL-6) signaling, are shown to play essential roles in the pathophysiology of the metabolic syndrome, type 2 diabetes, and subsequent cardiovascular diseases (1–3). As a causative factor of chronic inflammation, several lines of evidence support the role of free fatty acids. Of note, saturated fatty acids (SFAs), such as palmitate (Pal) and stearate, have been shown to induce proinflammatory cytokine production in various cell types, including macrophages (4,5). However, the underlying mechanism of SFA-induced proinflammatory cytokine expression is only partially elucidated.
To date, we and others have demonstrated that Toll-like receptor-4 (TLR4), a pathogen sensor expressed on the cell surface, plays a critical role in the SFA-induced proinflammatory cytokine expression (4–6). On the other hand, multiple mechanisms are involved in the SFA-induced cellular responses (1,7,8). Among them, attention has been focused on the role of cellular metabolic stresses such as endoplasmic reticulum (ER) stress and oxidative stress (1). Recent reports suggest that the modulation of metabolic stress pathways may alter high-fat diet–induced proinflammatory cytokine expression as well as insulin resistance (9,10). Therefore, it is of importance to clarify the molecular mechanism by which metabolic stresses affect proinflammatory cytokine expression.
In this study, using DNA microarray and network analyses in macrophages, we show that activating transcription factor (ATF) 4, a basic leucine zipper transcription factor, is potently induced by SFAs. We provide evidence that ATF4 plays essential roles in Il6 expression induced by various metabolic stresses, including ER stress. Furthermore, the ATF4 pathway has a synergistic effect on the TLR4 signaling pathway, enhancing Il6 expression. As a molecular mechanism, ATF4 is capable of enhancing metabolic stress–induced nuclear factor-κB (NF-κB) activation and directly activating the Il6 promoter. Our data suggest that ATF4 is a novel link between metabolic stress and Il6 expression in macrophages.
Research Design and Methods
Mice, Cells, and Reagents
Tlr4- and Atf4-deficient mice were provided by Dr. Shizuo Akira (Osaka University, Osaka, Japan). Mice were maintained on the C57BL/6 (B6) genetic background. All animal experiments were performed in accordance with the guidelines of Tokyo Medical and Dental University (no. 2011–207C, no. 0130269A). Murine peritoneal and bone marrow–derived macrophages (BMDMs) were prepared as described (11). All primary cells used for in vitro experiments were on the B6 background. Splenic CD11b–positive cells were purified using anti-CD11b magnetic beads and LS-columns (Miltenyi Biotec, Bergisch Gladbach, Germany). The RAW264 macrophage cell line (RIKEN BioResource Center, Tsukuba, Japan) was maintained in Dulbecco’s modified Eagle’s medium (Nacalai Tesque, Kyoto, Japan) containing 10% FBS. Palmitate (P5585), stearate (P4751), lipopolysaccharide (LPS; P4391), lipid A (L5399), and a double-stranded RNA–dependent protein kinase (PKR) inhibitor (2-aminopurine; A3509) were purchased from Sigma-Aldrich (St. Louis, MO). Fatty acids were solubilized in ethanol, and conjugated with fatty acid-free and Ig-free BSA at a molar ratio of 10:1 (fatty acid/albumin) in serum-poor medium (0.5% FBS) (4). The vehicle control used was a mixture of ethanol and BSA alone in place of fatty acids. A PKR-like ER kinase (PERK) inhibitor (GSK2606414) was obtained from EMD Millipore (Calbiochem; Billerica, MA). A stearoyl-CoA desaturase 1 inhibitor (A939572) was from BioVision (Milpitas, CA). Antibodies against ATF4 (sc200), RelA (sc372), ATF3 (sc188), β-actin (sc47778), and Lamin A/C (sc20681) were from Santa Cruz Biotechnology (Dallas, TX). Antibodies against IκBα (catalog #9242), eukaryotic initiation factor-2α (eIF2α; catalog #9722), and phospho-eIF2α (catalog #9721) were from Cell Signaling Technology (Danvers, MA). An antibody against ATF6 (IMG273) was from Imgenex (San Diego, CA).
Microarray, Network, and Pathway Analyses
Total RNAs were extracted from thioglycollate-elicited peritoneal macrophages using the RNeasyMini Kit (Qiagen, Valencia, CA). Microarray analysis was performed on biological duplicate samples using Affymetrix GeneChip Mouse Genome 430 2.0 Arrays according to the manufacturer’s instructions (12). Genes with an average fold change >1.8 and labeled as both “present (P)” and “increased (I)” were considered to be differentially upregulated. Affymetrix probe IDs were converted to unique Entrez IDs. Protein networks were built running the Markov clustering algorithm using the functional and physical interaction scores from STRING 9.0 with an inflation parameter of 2.0. Markov clustering has been used by many groups to build protein complexes or families based on protein-protein interactions (13,14). The results were visualized in Cytoscape software (15). Pathway and gene ontology analyses were performed using the Reactome functional protein interaction database (http://www.reactome.org/).
Nuclear Cytoplasmic Fractionation
The cells were collected in ice-cold PBS, resuspended in buffer A (10 mmol/L HEPES-KOH at pH 7.8, 10 mmol/L KCl, 0.1 mmol/L EDTA, 1 mmol/L dithiothreitol, 0.5 mmol/L phenylmethylsulfonyl fluoride, and protease inhibitors) and incubated on ice for 5 min. The cells were then spun down, and the cytoplasmic fraction was aspirated to separate tubes. The nuclear fraction was then lysed in buffer C (50 mmol/L HEPES-KOH at pH 7.8, 420 mmol/L KCl, 0.1 mmol/L EDTA, 5 mmol/L MgCl2, 1 mmol/L dithiothreitol, 0.5 mmol/L phenylmethylsulfonyl fluoride, and protease inhibitors) at 4°C for 30 min. The lysate was clarified by centrifugation, and the supernatant was collected.
Chromatin Immunoprecipitation Assay
Chromatin immunoprecipitation was performed using the MAGnify Chromatin Immunoprecipitation System (49–2024; Invitrogen, Carlsbad, CA) according to the manufacturer’s instructions with some modifications. In brief, thioglycollate-elicited peritoneal macrophages were cross-linked with 1% (w/v) formaldehyde for 10 min and lysed in the buffer provided. Nuclear extracts from 3 × 106 to 1 × 107 cells were used per immunoprecipitation reaction. Sonicated nuclear extracts were immunoprecipitated for 2 h at 4°C with anti-ATF4 (sc200x, 3 μg), anti-RelA (sc372, 3 μg), or IgG isotype negative control (sc2027, 3 μg) antibodies (Santa Cruz Biotechnology). DNA was eluted and purified as previously described (16). Eluted DNA was quantified with quantitative PCR using SYBR GREEN chemistry (StepOne Plus; Applied Biosystems, Foster City, CA).
RNA Purification, Reverse Transcription, and Real-Time PCR Amplification
RNA was purified using Sepasol (30486–56; Nacalai Tesque), according to the manufacturer’s instructions. Total RNA (1.25 μg) was reverse-transcribed using a ReverTraAce (FSQ-201; Toyobo, Osaka, Japan). Ten nanograms of cDNA were used for real-time PCR amplification with SYBR GREEN detection protocol in a thermal cycler (StepOne Plus; Applied Biosystems). Data were normalized to the 36B4 levels and analyzed using the comparative threshold cycle method (12). Primer sequences are listed in Supplementary Table 1.
Plasmid Construction, Transfection, and Knockdown Experiments
An expression vector encoding murine ATF4 was constructed by inserting PCR-amplified cDNA fragment encoding ATF4 between the EcoRI and BamHI sites of the pcDNA3.1Myc-His3 expression vector. For generation of the ATF4 5′ untranslated region (UTR) luciferase reporter plasmid, the 285 base pairs of murine ATF4 5′ UTR was amplified by PCR and was inserted into the HindIII site of the pGL3 control vector (E1741; Promega, Madison, WI). The rat Il6 promoter luciferase reporter plasmid and various truncated constructs (17) were gifts from Dr. Toshihiro Ichiki (Kyushu University, Fukuoka, Japan). The Il6 promoter construct containing mutant ATF/cAMP-responsive element–binding protein site was generated from the Il6 promoter construct 2 by the inverse PCR-based site-directed mutagenesis using KOD plus DNA polymerase (KOD201; Toyobo). The thymidine-kinase promoter renilla luciferase reporter vector was purchased from Promega. All transfection experiments were performed using Lipofectamine LTX reagent (A12621; Invitrogen) according to the manufacturer’s protocol. Retrovirus-mediated knockdown of ATF4 was performed as previously described (18). The target sequences for short hairpin (sh) ATF4 were 5′-TCCCTCCATGTGTAAAGGA-3′ (shATF4 1) and 5′-CTCTGTTTCGAATGGATGA-3′ (shATF4 2), respectively. As a negative control, shGFP was used as described previously (18).
For the ATF4 5′ UTR luciferase assay, RAW264 macrophages seeded on 24-well plates were transiently cotransfected with 30 ng renilla and 600 ng firefly luciferase reporter plasmids. For the Il6 promoter luciferase assay RAW264 macrophages were transiently cotransfected with 50 ng renilla and 250 ng firefly luciferase reporter plasmids. Twenty-four hours after transfection, cells were collected and luciferase activity was measured with a dual-luciferase reporter assay system (Promega) according to the manufacturer’s protocol. All data were normalized for transfection efficiency by the division of firefly luciferase activity by renilla luciferase activity.
NF-κB RelA DNA-Binding Activity Assay
Ten micrograms of nuclear extracts was used to determine RelA (p65) DNA–binding activity using an ELISA-based assay (TransAM 40096; Active Motif, Carlsbad, CA), according to the manufacturer’s instructions. In brief, κB oligonucleotide–coated plates (in a 96-well format) were incubated for 1 h with the nuclear extracts. Specificity was achieved by incubation with anti-RelA primary antibodies for 1 h. Horseradish peroxidase–conjugated secondary antibodies were used for the detection of RelA bound to the κB sequences.
ELISA of mouse IL-6 was performed using Mouse IL-6 Quantikine ELISA Kit (R&D Systems, Minneapolis, MN), as described previously (19).
Data were expressed as the mean ± SEM. P values were calculated using two-tailed Student t tests.
ATF4 Pathway Is Activated by SFAs in the Absence of TLR4
In an attempt to identify novel mechanisms underlying the SFA-induced inflammatory responses, we performed DNA microarray analysis of palmitate-stimulated peritoneal macrophages obtained from Tlr4-deficient and wild-type mice. A total of 122 genes were upregulated by palmitate in both Tlr4-deficient and wild-type macrophages (Fig. 1A). We next performed Markov clustering of these genes using STRING database (20), which produced several clusters (Fig. 1B). Pathway analysis using the Reactome database (21) identified two large clusters in which target genes for ATF4 and NF-κB were enriched (Fig. 1C). Indeed, several signaling pathways that can induce ATF4 were significantly enriched in gene ontology analysis (Supplementary Fig. 1).
ATF4 is induced following phosphorylation of eIF2α, which is mediated by eIF2α kinases (i.e., the heme-regulated eIF2α kinase, PKR, PERK, and the general control nonderepressible 2) under metabolic stresses such as ER stress, oxidative stress, and amino acid deprivation (22). Previous reports showed that at least two of the eIF2α kinases, PKR and PERK, are activated by SFAs (9,23). Consistently, pharmacological inhibitors of PKR and PERK suppressed palmitate-induced activation of the ATF4 pathway (Supplementary Fig. 2A and B), suggesting that PKR and PERK are likely to be involved in SFA-induced eIF2α phosphorylation. When eIF2α is phosphorylated, ATF4 translation is preferentially upregulated through the characteristic upstream open reading frames in the 5′ UTR (24). Using ATF4 5′ UTR luciferase assay, we confirmed that ATF4 translation was activated by palmitate in RAW264 macrophage–like cell lines (Fig. 1D). In line with this, palmitate increased ATF4 protein levels and the expression of target genes, such as tribbles homolog-3 (Trib3) and c/ebp homologous protein (Chop, also known as Ddit3) (25), in peritoneal macrophages, at least in part, independent of TLR4 (Fig. 1E, and Supplementary Fig. 2D). We also observed that other branches of the unfolded protein response were activated by palmitate in the absence of TLR4 (Supplementary Fig. 2C and D). Intriguingly, we found that, even in Tlr4-deficient macrophages, palmitate increased the expression of some proinflammatory cytokines, such as Il6 and tumor necrosis factor (Tnf), to a lesser extent than wild-type macrophages (Supplementary Fig. 2D). Consistent with this, nuclear protein levels and activity of RelA (p65), a transcriptional activating subunit of NF-κB, were increased by palmitate in both wild-type and Tlr4-deficient macrophages (Fig. 1E and F). These results suggest that the ATF4 pathway, as well as the NF-κB pathway, is activated in macrophages by palmitate in the absence of TLR4.
In addition, we confirmed that stearate, another SFA, also activated the ATF4 pathway in RAW264 macrophages (Supplementary Fig. 3A). By contrast, unsaturated fatty acids such as oleate and eicosapentaenoic acid did not activate the ATF4 pathway (Supplementary Fig. 3B). Notably, eicosapentaenoic acid effectively suppressed palmitate-induced activation of the ATF4 pathway (Supplementary Fig. 3B). We also found that pharmacological inhibition of fatty acid desaturation resulted in upregulation of ATF4 target genes (Supplementary Fig. 3C) (26,27), suggesting that intracellular SFAs, not unsaturated fatty acids, play a role in the ATF4 pathway activation. Although further studies are required, our results are consistent with previous reports showing that altered membrane lipid composition may activate unfolded protein response pathways through several mechanisms, including calcium depletion (28) and direct activation of ER membrane sensors (29).
ATF4 Is Required for Il6 Expression in Response to Metabolic Stresses
In contrast to NF-κB (30), the role of ATF4 in the inflammatory pathway is not yet fully characterized. Because Atf4-null mice are mostly embryonic or neonatal lethal (31), in this study we used primary macrophages obtained from Atf4-haploinsufficient mice. We found that BMDMs from Atf4-haploinsufficient mice were defective in palmitate-induced mRNA expression of Il6 (Fig. 2A). Tnf expression was only marginally attenuated in Atf4-haploinsufficient macrophages. Consistently, Atf4-haploinsufficient macrophages showed decreased secretion of IL-6 in the media (Fig. 2B). Similarly, Atf4-haploinsufficient peritoneal macrophages showed attenuated Il6 expression induced by ER stressors, such as tunicamycin (Fig. 2C and D) and thapsigargin (Fig. 2E), compared with wild-type macrophages. To test the importance of ATF4 in RAW264 macrophages, we knocked down Atf4, which was confirmed by quantitative PCR and Western blotting (Supplementary Fig. 4A and B). Atf4 knockdown reduced SFA-induced (Supplementary Fig. 4C) or ER stress–induced (Supplementary Fig. 4D) Il6 expression. On the other hand, in macrophages treated without these stressors, there was no significant difference in lipid A (a bona fide TLR4 ligand)–induced proinflammatory cytokine expression between the genotypes (Fig. 2F). To test the in vivo functional role of ATF4, we activated the ATF4 pathway by intraperitoneal injection of tunicamycin (Fig. 2G). Sixteen hours after the injection, tunicamycin treatment potently induced Atf4 and Il6 mRNA expression in the spleen and liver from wild-type mice, which was significantly decreased in those from Atf4-haploinsufficient mice. Tnf expression was not significantly decreased. We obtained a similar result using purified CD11b-positive splenic macrophages (Supplementary Fig. 5). In this study, we observed no apparent difference in LPS-induced Il6 mRNA expression between the genotypes (data not shown). These observations suggest that ATF4 plays a critical role in the metabolic stress–induced proinflammatory cytokine expression in vitro and in vivo.
ATF4 Pathway Enhances Il6 Expression in Synergy With the TLR4 Pathway
We next examined the impact of the ATF4 pathway activation on proinflammatory cytokine expression in cells under metabolic stresses. Pretreatment of various macrophages with ER stressors markedly enhanced the lipid A–induced mRNA expression and secretion of IL-6 (Fig. 3A–C and Supplementary Fig. 6A). The expression levels of Tnf were also affected to a lesser extent (Fig. 3A and B). Similar results were obtained using murine embryonic fibroblasts (MEFs) (Supplementary Fig. 6C). Although the magnitude varied, the patterns of gene expression in response to ER stressors and TLR4 agonists were similar among cell types. Notably, this enhancement was significantly attenuated in Atf4-haploinsufficient macrophages (Fig. 3D), ATF4 knockdown RAW264 macrophages (Supplementary Fig. 6B), and Atf4-deficient MEFs (Supplementary Fig. 6C). These observations suggest that the ATF4 pathway has a synergistic effect on the TLR4 signaling pathway.
In addition to the eIF2α-ATF4 branch, ER stress activates two other branches of the unfolded protein response: the inositol requiring enzyme 1α X-box binding protein 1 (XBP1) and ATF6 branches, the former of which is also involved in proinflammatory cytokine expression (32). To specifically stimulate the ATF4 pathway, we used salubrinal (Sal), an inhibitor of eIF2α phosphatase complex (33). Pretreatment of BMDMs with Sal resulted in upregulation of ATF4 (Fig. 3E) and its target genes, whereas its effect on XBP1 splicing was minimal (Fig. 3F). Moreover, pretreatment with Sal potently enhanced the lipid A–induced mRNA expression and secretion of IL-6 in BMDMs (Fig. 3G and H) and RAW264 macrophages (Supplementary Fig. 6D and E). We also examined the effect of amino acid deprivation, another metabolic stress known to activate the ATF4 pathway (34), on Il6 expression. Pretreatment with azetidine, a proline analog, significantly enhanced the lipid A–induced Il6 expression (Supplementary Fig. 6F). Transient overexpression of ATF4 showed similar effects in RAW264 macrophages (Fig. 3I and J). These observations confirm that, in addition to the previously characterized XBP1 pathway, the ATF4 pathway is also involved in the regulation of metabolic stress–induced proinflammatory cytokine expression. Of note, a recent report showed that the ATF4 pathway was attenuated following the low-dose treatment of TLR4 ligands in cultured macrophages (34). Collectively, it is conceivable that there is a bidirectional crosstalk between the ATF4 and TLR4 pathways.
ATF4 Is Involved in NF-κB Activation in Response to Metabolic Stresses
We next aimed to clarify the molecular mechanism of the SFA-induced Il6 expression through the ATF4 pathway. First, we examined the effect of ATF4 on palmitate-induced NF-κB activation, because our microarray analysis revealed that the NF-κB pathway was activated in Tlr4-deficient macrophages (Fig. 1B and C). In this study, we found that the nuclear protein levels and activity of RelA were increased by palmitate even in the absence of TLR4 (Fig. 1E and F). Interestingly, Atf4 haploinsufficiency markedly reduced nuclear protein levels of RelA induced by palmitate or thapsigargin in peritoneal macrophages (Fig. 4A and B). We next examined the kinetics of cytoplasmic IκBα degradation, a well-recognized event prior to nuclear translocation of RelA (30), and found no apparent difference in cytoplasmic IκBα protein levels between the genotypes (Fig. 4C). Moreover, we observed that NF-κB activity induced by palmitate was significantly attenuated in Atf4-haploinsufficient macrophages relative to wild-type macrophages (Fig. 4D). Consistently, a chromatin immunoprecipitation assay showed decreased recruitment of RelA to the Il6 promoter region in Atf4-haploinsufficient macrophages (Fig. 4E). Because the transcriptional activity of RelA can be regulated by multiple mechanisms (30), further studies are required to elucidate how ATF4 affects RelA activity. Although a previous report showed that phosphorylation of eIF2α leads to NF-κB activation in MEFs (35), our data provide novel evidence that ATF4 is involved in metabolic stress–induced NF-κB activation in macrophages.
ATF4 Directly Activates the Il6 Promoter
On the basis of its potent proinflammatory effect, we next examined whether ATF4 directly activates Il6 transcription. The consensus binding sequence of ATF4 is identical to that of the cAMP response element (CRE) (36). The Il6 promoter contains an evolutionarily conserved region in which a putative ATF4-binding sequence is located in the vicinity of the NF-κB–binding site (Fig. 5A). We, therefore, performed the Il6 promoter luciferase assay using RAW264 macrophages. The luciferase activity was increased by treatment with LPS, which was further enhanced by ATF4 overexpression (Fig. 5B). The effect of ATF4 overexpression was abolished when transfected with the promoter construct containing a mutant CRE. Interestingly, the mutation in the CRE also markedly inhibited the LPS-induced activation of the Il6 promoter. Consistently, experiments with a series of truncated Il6 promoter constructs showed marked reduction of the promoter activity in the absence of CRE (Supplementary Fig. 5), suggesting that the CRE is indispensable for the LPS- and ATF4-induced activation of the Il6 promoter. To confirm the direct recruitment of ATF4 to the Il6 promoter region, we performed a chromatin immunoprecipitation assay using peritoneal macrophages. Signals from the promoter region containing the CRE were significantly increased upon treatment with palmitate or thapsigargin plus lipid A (Fig. 5C). These data suggest that ATF4 directly activates the Il6 promoter.
In this study, we demonstrated that ATF4 is induced downstream of metabolic stresses caused by SFAs and ER stressors. ATF4 exerts its proinflammatory effects through at least two different mechanisms: direct activation of the Il6 promoter and involvement in NF-κB activation (Fig. 5D). According to previous reports, other members of ATF/CREB family of transcription factors are also involved in positive (e.g., XBP1) or negative (e.g., ATF3) regulation of Il6 expression (18,32). Of note, these factors may be induced downstream of TLR4 stimulation even independent of ER stress (18,32). By contrast, our results suggest that ATF4 is minimally involved in the TLR4 signaling in unstressed cells. Considering that the ATF4 pathway is activated under a variety of metabolic stresses, ATF4 would constitute a critical link between metabolic stresses and Il6 expression.
Furthermore, our observations that the ATF4 pathway acts in synergy with the TLR4 pathway raise a possibility that metabolic stresses affect innate immune response. Consistent with several previous reports (4–6), our results suggest that TLR4 is required for proinflammatory effects of SFAs. Importantly, the current study also identifies the ATF4 pathway as a novel mechanism of SFA-induced proinflammatory cytokine expression that is induced even in the absence of TLR4. Given that SFAs activate both TLR4 and various metabolic stress pathways upstream of ATF4 (1,4–9), the cross talk between these pathways would be important for better understanding of the molecular mechanism of SFA-induced proinflammatory cytokine expression. Collectively, our findings raise the possibility that ATF4 plays a role in the pathophysiology of chronic inflammation in the metabolic syndrome and type 2 diabetes.
See accompanying commentary, p. 48.
Acknowledgments. The authors thank Dr. Shizuo Akira, Osaka University, for providing Atf4 haploinsufficient mice; Dr. Toshihiro Ichiki, Kyushu University, for providing Il6 promoter luciferase plasmids; and Dr. Takahisa Nakamura, Cincinnati Children's Hospital Medical Center, for critical reading of the manuscript.
Funding. This work was supported in part by Grants-in-Aid for Scientific Research from the Ministry of Education, Culture, Sports, Science and Technology of Japan; Japan Science and Technology Agency, Precursory Research for Embryonic Science and Technology; Astellas Foundation for Research on Metabolic Disorders; and Nestlé Nutrition Council, Japan.
Duality of Interest. No potential conflicts of interest relevant to this article were reported.
Author Contributions. Y.I. and T.S. researched data, contributed to the discussion, and wrote, reviewed, and edited the manuscript. R.H., M.K.-S., M.T., and M.H. contributed to the discussion. I.S. researched data and contributed to the discussion. T.T.-I. contributed to network and pathway analyses. M.N. and Y.M. performed the microarray experiments. Y.O. contributed to the discussion and wrote, reviewed, and edited the manuscript. T.S. and Y.O. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.