Plasma glutamate–modulated interaction of A_2AR and mGluR5 on BMDCs aggravates traumatic brain injury–induced acute lung injury

In the severe TBI-induced ALI model, we found that selective inactivation of BMDC A_2AR (KO→WT) significantly reduced lung water content (Fig. 1 A), elevated PaO₂/FIO₂ (P/F) ratios (Fig. 1 B), and attenuated histological signs of pulmonary injury (Fig. 1 C) at 24 h after injury when compared with WT littermates, consistent with the results observed in global A_2AR KO mice. Conversely, in the oleic acid–induced ALI model, each of these measures of lung damage was exacerbated in the mice with selective inactivation of BMDC A_2AR or global A_2AR KO (Fig. 1). Reconstitution of BMDC A_2AR in global A_2AR KO mice (WT→KO) eliminated the A_2AR KO-triggered protective effects in TBI-induced ALI and the deleterious effects in oleic acid–induced ALI. In addition, no significant difference was observed between WT mice and WT mice transplanted with WT BMDCs (WT→WT) or between KO and KO→KO transplanted mice, excluding the possibility that these results were induced nonspecifically by BM transplantation (BMT; Fig. 1). These data strongly suggest that in contrast to the protective effect of BMDC A_2AR in oleic acid–induced ALI, BMDC A_2AR plays a deleterious role in TBI-induced ALI.

As shown in Fig. 2 A, at 24 h after TBI, global A_2AR KO mice and mice with selective inactivation of BMDC A_2AR (KO→WT) displayed reduced numbers of pulmonary neutrophils for TBI-induced ALI but enhanced numbers for oleic acid–induced ALI. Similarly, global or BMDC-specific A_2AR KO mice showed increased pulmonary levels of the proinflammatory cytokines TNF and IL-1β after TBI-induced ALI and reduced levels after oleic acid–induced TBI (Fig. 2 B). Reconstitution of BMDC A_2AR in global A_2AR KO mice (WT→KO) eliminated the A_2AR KO–triggered antiinflammatory effect in TBI-induced ALI and the proinflammatory effect in oleic acid–induced ALI (Fig. 2). These results indicate that BMDC A_2AR plays a significant proinflammatory role in the progression of TBI-induced ALI.

In both the TBI- and oleic acid–induced ALI models, we measured plasma glutamate levels and found that they were much higher in mice with severe TBI-induced ALI (25–50 µmol/l) than in mice with oleic acid–induced ALI (8–15 µmol/l) at 6, 12, and 24 h after injury, even though the plasma glutamate level was elevated in both models (Fig. 3 A). To corroborate this, we also assayed the changes in plasma glutamate levels in patients suffering from TBI-induced ALI or lung trauma–induced ALI, which is also a nonneurogenic ALI. Consistent with the results in the animal models, the plasma glutamate concentrations of patients suffering from severe TBI-induced ALI were much higher than those of patients with lung trauma–induced ALI at 6, 12, and 24 h after injury (Fig. 3 B).

Glutamate has been previously reported to regulate A_2AR activity in brain injury, and heteromeric complexes of adenosine receptors and mGluRs were found in the brain under physiological or pathological conditions (Ciruela et al., 2001; Ferré et al., 2002; Nishi et al., 2003; Dai et al., 2010b). However, only mGluR5 and mGluR2/3 have been reported to be expressed on white blood inflammatory cells (Gill and Pulido, 2001; Liu et al., 2010b). Accordingly, using specific antibodies for A_2AR, mGluR5, and mGluR2/3, we performed coimmunoprecipitation (co-IP) experiment to investigate the interaction between A_2AR and these mGluRs. From WBC extracts of mice with TBI-induced ALI, the mouse antibody against A_2AR coimmunoprecipitated a band corresponding to mGluR5 (Fig. 4 A). This band did not appear in immunoprecipitates from the cells of mice with oleic acid–induced ALI (Fig. 4 A). No IP of mGluR2/3 was obtained from WBCs in either of the two mouse ALI models (Fig. 4 A). Consistent with these results in animal models, analysis by SDS-PAGE and immunoblotting with anti-mGluR5 antibodies of immunoprecipitates obtained with human anti-A_2AR showed a band corresponding to mGluR5 in WBCs of TBI-induced ALI patients but not those of lung trauma–induced ALI patients (Fig. 4 B). No mGluR2/3 band was detected in WBCs of either group (Fig. 4 A). In addition, using anti-mGluR5 antibody as the immunoprecipitating antibody and anti-A_2AR antibody as the immunoblotting antibody, the interaction of A_2AR and mGluR5 was confirmed in the WBCs from mice and patients (Fig. 4, C and D). These results indicate that A_2AR–mGluR5 interaction is induced in BMDCs in severe TBI-induced ALI.

To confirm the links between plasma glutamate level, A_2AR–mGluR5 interaction in BMDCs, and TBI-induced lung damage, we used a mouse model of moderate TBI. Of 40 experimental mice with moderate TBI, 38 displayed decreased plasma glutamate levels at 6 h after TBI and recovered almost to baseline at 12 h after injury, which did not exhibit ALI after moderate TBI. Only two mice displayed increased plasma glutamate levels after injury (Fig. 5 A). At 24 h after TBI, only these two mice suffered from ALI with increased lung water content (Fig. 5 B), pathological changes in lung tissues (Fig. 5 D), and deficiency of pulmonary gas exchange (P/F ratio was below 300 mmHg; Fig. 5 C). Moreover, A_2AR–mGluR5 interaction in WBCs was observed only in these two mice (Fig. 5, E and F).

To explore whether the elevated plasma glutamate level in TBI-induced ALI is involved in inducing the A_2AR–mGluR5 interaction and whether this interaction mediates the proinflammatory effect of BMDC A_2AR, we investigated the receptor interaction and the induction of proinflammatory cytokines in human and mouse neutrophils in vitro in the presence of varying concentrations of glutamate (the given glutamate concentrations were relative to the plasma glutamate levels in vivo described above). In mouse neutrophils in the presence of 30 and 300 µmol/l glutamate, a band corresponding to mGluR5 was coimmunoprecipitated by anti-A_2AR antibodies (Fig. 6 A), and an A_2AR band was coimmunoprecipitated by anti-mGluR5 antibodies (Fig. 6 B), indicating the interaction between these receptors at these glutamate concentrations. At lower concentrations of glutamate, no co-IP was observed (Fig. 6, A and B). mGluR5 siRNA significantly decreased the expression of mGluR5 in mouse neutrophils, establishing our ability to knock down mGluR5 (Fig. 6 C). As shown in Fig. 6 D, the A_2AR agonist CGS21680 inhibited LPS-induced IL-1β mRNA expression in neutrophils from WT mice in control medium and in the presence of 5 µmol/l glutamate but enhanced IL-1β mRNA levels in the presence of 30 and 300 µmol/l glutamate (Fig. 6 D). This glutamate concentration–dependent change was not observed in neutrophils from A_2AR KO mice or when mGluR5 was knocked down by mGluR5 siRNA in WT neutrophils (Fig. 6 D).

Similar results were obtained in human neutrophils: in the presence of 300 µmol/l glutamate (similar to the plasma glutamate level in patients with TBI-induced ALI), an mGluR5 band was coimmunoprecipitated by anti-A_2AR antibody (Fig. 6 E) and an A_2AR band was coimmunoprecipitated by anti-mGluR5 antibody (Fig. 6 F). And no co-IP was observed at lower concentrations of glutamate (Fig. 6, E and F). Moreover, CGS21680 switched from inhibiting (in the presence of 0, 5, or 30 µmol/l glutamate) to promoting (at 300 µmol/l glutamate) LPS-induced IL-1β mRNA expression in human neutrophils, dependent on glutamate concentration (Fig. 6 G). The antiinflammatory effect of CGS21680 in the presence of 0, 5, and 30 µmol/l glutamate was completely blocked by the selective A_2AR antagonist ZM241385, whereas the proinflammatory effect of CGS21680 in the presence of 300 µmol/l glutamate was suppressed to some extent by ZM241385 or mGluR5 antagonist 2-methyl-6-(phenylethynyl)pyridine (MPEP) alone but completely suppressed by a combination of ZM241385 and MPEP (Fig. 6 G). These findings suggest that the increased plasma glutamate level caused by TBI-induced ALI promotes the interaction between A_2AR and mGluR5 in BMDCs and that this interaction is critical for the proinflammatory effect exerted by activation of BMDC A_2AR.

We next investigated the signaling pathway associated with the A_2AR–mGluR5 interaction–associated proinflammatory effect. As shown in Fig. 7 A, the stimulatory effect of CGS21680 on LPS-induced IL-1β mRNA expression in mouse neutrophils in the presence of 30 and 300 µmol/l glutamate was suppressed by the PKC inhibitor GF109203X but not by the PKA inhibitor H-89. Consistent with this, Western blots showed increased PLC expression in the presence of 30 or 300 µmol/l glutamate, whereas adenylyl cyclase (AC) expression was higher at lower glutamate concentrations (Fig. 7 B). However, in neutrophils from A_2AR KO mice or neutrophils pretreated with mGluR5 siRNA, this glutamate level–dependent change in expression of signaling molecules was not observed (Fig. 7 B). We obtained similar results in human neutrophils: H-89 blocked the antiinflammatory effect of A_2AR activation in the presence of 0, 5, and 30 µmol/l glutamate but had no effect on its proinflammatory effect in the presence of 300 µmol/l glutamate, which was instead inhibited by GF109203X (Fig. 7 C). The increase in AC expression detected at lower glutamate concentrations was blocked by the A_2AR antagonist ZM241385, whereas the significantly elevated PLC levels detected in the presence of high concentrations of glutamate were largely inhibited by a combination of ZM241385 and the mGluR5 antagonist MPEP (Fig. 7 D). These data indicate that the A_2AR–mGluR5 complex–dependent proinflammatory effect of A_2AR activation that occurs in the presence of elevated glutamate levels is mediated by a PLC–PKC signaling pathway.

To verify the plasma glutamate–modulated proinflammatory role of BMDC A_2AR described above, we investigated the effects of A_2AR agonist and antagonist on lung damage in a mouse model of severe TBI-induced ALI. As shown in Fig. 8, in WT mice suffering from severe TBI-induced ALI, the A_2AR agonist CGS21680 aggravated lung damage as measured by lung water content (Fig. 8 A), P/F ratio (Fig. 8 B), pulmonary pathological changes (Fig. 8 C), and neutrophil infiltration into lung tissues (Fig. 8 D). The A_2AR antagonist ZM241385 attenuated the lung damage as measured by these parameters (Figs. 8, A–D). These effects of CGS21680 and ZM241385 were not observed in A_2AR KO mice (Fig. 8). These data confirm the aggravating effect of A_2AR activation in severe TBI-induced ALI and suggest that A_2AR antagonist is protective in the treatment of neurogenic ALI.

Inflammatory response plays an important role in the progression of ALI. Activation of A_2AR is widely considered to be an antiinflammatory event in nonneurogenic ALI, as A_2AR inhibits BMDC activation and inflammatory cytokine release. Consistent with this perspective, our findings in an oleic acid–induced ALI model show that global A_2AR KO and selective inactivation of BMDC A_2AR accelerate lung inflammatory damage. In contrast, in the severe TBI-induced ALI mouse model, selective inactivation of BMDC A_2AR significantly attenuated lung damage and expression of inflammatory cytokines TNF and IL-1β. This result provides strong evidence that BMDC A_2AR activation has opposite effects on nonneurogenic ALI and the neurogenic ALI induced by severe TBI.

Our previous study in mouse models of moderate TBI and in isolated microglia demonstrated that high concentrations of glutamate in the environment can switch the antiinflammatory effect of A_2AR activation to a proinflammatory effect (Dai et al., 2010b). This prompted us to measure glutamate levels in blood plasma during BMDC A_2AR activation and to investigate whether glutamate levels also play a role in the progression of TBI-induced ALI. Our results show that both in patients with severe TBI-induced ALI and in mouse models of TBI-induced ALI, plasma glutamate levels during the 24 h after injury were much higher than in lung trauma–induced ALI patients or in mice with oleic acid–induced ALI. Moreover, we found a significant A_2AR–mGluR5 interaction in WBCs from patients with severe TBI-induced ALI and WBCs from the mouse model of severe TBI-induced ALI but not in WBCs from patients with trauma-induced ALI or the oleic acid–induced ALI mouse model. This association between plasma glutamate level, lung injury, and BMDC A_2AR–mGluR5 interaction was confirmed by the investigation in the mouse model of moderate TBI. In the very small minority of mice suffering from ALI after moderate TBI, we also detected elevated plasma glutamate levels, lung damage, and A_2AR–mGluR5 interaction in WBCs. Numerous experimental and clinical studies have demonstrated that CSF or extracellular fluid glutamate concentration increases significantly after TBI, reflecting neuronal and glial glutamate release, diminished glial glutamate uptake, and a damaged blood–brain barrier (Baker et al., 1993; Palmer et al., 1993, 1994; Tomkins et al., 2008). However, further study is required to determine whether the deficiency of glutamate transporter, insufficient capacity of the blood glutamate scavenger oxaloacetate, muscle integrity and muscle breakdown (Lee et al., 1998; O’Kane et al., 1999; Hyde et al., 2003; Hawkins et al., 2006; Zlotnik et al., 2010), or some other mechanism accounts for the increase of plasma glutamate we observed in TBI-induced ALI in the present study.

Previously, functional A_2AR–mGluR5 heteromeric complexes had only been reported in the central nervous system (Ferré et al., 2002; Nishi et al., 2003; Brown et al., 2012). The present study provides the first evidence for this interaction in BMDCs and the first evidence that the complex is induced in ALI. Infiltration of neutrophils associated with inflammatory cytokine release is a hallmark event in the progression of ALI and its severe form, acute respiratory distress syndrome (Baughman et al., 1996; Abraham et al., 2000; Azoulay et al., 2002; Reutershan and Ley, 2004). To confirm that TBI-induced high plasma glutamate levels trigger the A_2AR–mGluR5 interaction and that the complex is an important mediator of the proinflammatory effect of BMDC A_2AR, we performed additional experiments in isolated human and mouse neutrophils. These in vitro experiments revealed that elevated glutamate levels (300 µmol/l for human neutrophils; 30 and 300 µmol/l for mouse neutrophils) induced the A_2AR–mGluR5 interaction. The glutamate levels required to trigger the A_2AR–mGluR5 interaction in neutrophils are similar to those detected in patients with TBI-induced ALI and in the mouse model, respectively. In elevated glutamate environments, neutrophil A_2AR activation exerted a proinflammatory effect rather than the antiinflammatory effect observed in the presence of lower glutamate concentrations. Moreover, this effect was largely suppressed by either A_2AR antagonist ZM241385 or mGluR5 antagonist MPEP and completely blocked by their combination, confirming that A_2AR–mGluR5 interaction mediated this proinflammatory effect.

Both A_2AR and mGluR5 are G protein–coupled receptors (GPCRs) comprised of a bundle of seven transmembrane (7TM) α-helices connected by three extracellular loops and three intracellular loops (Katritch et al., 2012). Some previous studies have reported that agonists of GPCRs can affect the formation and stability of GPCR homo- or heterocomplexes (Hebert et al., 1996; Angers et al., 2000; Rocheville et al., 2000), provoke changes in downstream signaling, and regulate their functions (Jordan and Devi, 1999; Gomes et al., 2001; Rios et al., 2001; Cheng et al., 2003). These reports provide an explanation for our results. First, A_2AR–mGluR5 interaction and this interaction-mediated proinflammatory effect require both A_2AR agonist CGS21680 and the presence of a certain concentration of glutamate, the endogenous agonist of mGluR5. Second, this proinflammatory effect is mediated by PLC–PKC signaling rather than by the AC–PKA pathway responsible for the antiinflammatory effect of A_2AR activation in low concentrations of glutamate. It is believed that mGluR5 is coupled with Gq, which classically activates the PLC–PKC pathway (Ozawa et al., 1998). A_2AR is coupled with Gs, although AC-cAMP-PKA activation is the predominant downstream signaling pathway, provoking PKA-dependent or -independent PKC signaling (Pinto-Duarte et al., 2005; Fredholm et al., 2007). In the present study, we demonstrate that the PLC–PKC signaling pathway contributes to the glutamate-induced proinflammatory effect of A_2AR–mGluR5 interaction in a PKA-independent manner, which accounts for the deleterious effect of BMDC A_2AR in severe TBI-induced ALI. Confirming this finding, A_2AR antagonist rather than agonist attenuated lung inflammatory damages in severe TBI-induced ALI. Previously, injection of enough glutamate was reported to cause excitotoxic ALI by activating NMDA receptors in lung tissue (Said et al., 1996; Shen et al., 2007), but our results suggest that plasma glutamate–modulated interaction of A_2AR and mGluR5 on BMDCs aggravates TBI-induced ALI. It not only presents a novel view of the effect of A_2AR in ALI, but also of the role of glutamate in ALI.

In summary, we demonstrate that high levels of glutamate induce the A_2AR–mGluR5 interaction in BMDCs to trigger a proinflammatory effect of A_2AR activation via PLC–PKC signaling. This finding offers new insight into the clinical use of A_2AR modulators (agonist or antagonist) and glutamate inhibitor for TBI-induced ALI treatment. When plasma glutamate levels are elevated, A_2AR antagonist rather than agonist is a potentially beneficial treatment for attenuating lung damage in severe TBI-induced ALI. Importantly, A_2AR agonist may actually exacerbate the damage in these conditions.

Global A_2AR homozygous KO mice and their WT littermates used in this study were developed in the laboratory of J.-F. Chen (Chen et al., 1999) and were generated as previously described (Chen et al., 1999; Dai et al., 2010a,b). In brief, congenic global A_2AR KO mice on a C57BL/6 background were generated by backcrossing global A_2AR KO on mixed (129-Steel×C57BL/6) genetic background to C57BL/6 mice for 13–15 generations. Heterozygous crossbreeding was used to generate global A_2AR KO mice and their WT littermates. All procedures used in this study were approved by the Institutional Animal Care and Use Committee of the Third Military Medical University.

ALI is defined as an acute noncardiogenic pulmonary edema with a P/F gradient below 300 mmHg. Hospitalized patients with TBI-induced ALI or lung trauma–induced ALI and their blood cell cultures were eligible for investigation. This study was approved by the Third Military Medical University Human Studies Committee. The clinical information for these patients is presented in Tables S1 and S2.

To selectively inactivate or reconstitute A_2AR on BMDCs, BMT between WT mice and A_2AR KO mice was performed as previously described (Yu et al., 2004; Dai et al., 2010a).

A severe cortical impact was performed by the weight-dropping method (Li et al., 2008, 2009). This reproducible and consistent model is generally associated with 30% mortality within the first 5 min after injury and neurological deficit scores of III within the 24 h after injury (Jin et al., 2008). As in patients, ALI in mice was defined as lung edema with a P/F gradient below 300 mmHg, ruling out heart failure as a mechanism. The nonneurogenic ALI model was induced by intravenous injection of 0.6 ml/kg oleic acid (Sigma-Aldrich). In addition, to confirm the effect of TBI severity on lung damage, a moderate TBI model in mice was also induced via the weight-dropping method. This model yielded a neurological deficit score of II within the 24 h after injury. To investigate the effect of A_2AR agonist and antagonist in the mouse severe TBI-induced ALI model, A_2AR agonist CGS21680 (0.1 mg/kg) or antagonist ZM241385 (1 mg/kg) was given to the mice at 3 h after TBI.

The lung injury parameters of mice such as lung water content, hematoxylin and eosin (H&E) staining, and TNF and IL-1β protein levels were assessed as previously described (Liu et al., 2010a). Using goat anti–mouse A_2AR antibody (Everest Biotech Ltd.), rabbit anti–mouse CD177 antibody (Beijing Biosynthesis Biotechnology Co., Ltd.), and FITC- or TRITC-conjugated secondary antibody (Beijing Cowin Biotech Co., Ltd.), A_2AR location and neutrophil infiltration in lung tissue were determined with standard immunofluorescence immunohistochemistry (Donadieu et al., 2007). Nuclei were then stained by DAPI. The results were analyzed by Image-Pro plus version 4.5 (Media Cybernetics) as described previously (Li et al., 2008). Arterial blood samples were withdrawn into a heparinized syringe by percutaneous left ventricular sampling of lightly anesthetized mice spontaneously breathing room air (Fagan et al., 1999; Xu et al., 2006). Blood gas analysis was immediately performed using the I-STAT Analyzer (Abbott).

Glutamate levels in blood plasma of the patients (Tables S1 and S2) or mice were analyzed by HPLC using ortho-phthaldialdehyde precolumn derivatization and fluorescence detection as previously reported (Aliprandi et al., 2005). Samples were run through a C18 column (Supelcosil) detected by a 157 spectrofluorometric detector (Beckman Coulter) with the excitation wavelength set at 330 nm and the emission wavelength fixed at 450 nm.

Blood samples from patients or the mouse models were collected in EDTA tubes, and WBCs were prepared by isolating the buffy coat layer after centrifugation at 3,500 rpm for 10 min at 4°C. For the co-IP experiment, WBC proteins were extracted with TRIzol reagent (Invitrogen) from the interphase and lower phenol phase after first removing the upper aqueous phase according to the manufacturer’s instructions (Chung et al., 2009).

Using a discontinuous Percoll gradient (Henson and Oades, 1975; Graham et al., 2007), human neutrophils from healthy donors were isolated from heparinized blood, whereas mouse neutrophils were isolated from BM cells according to the method described previously (Tsai and Collins, 1993; Magalhães et al., 2007). Neutrophils were pretreated with 0, 5, 30, or 300 µM glutamate for 30 min, followed by a combination of 1,000 ng/ml LPS and 100 nM A_2AR agonist CGS21680 for 6 h (Dai et al., 2010b). To elucidate the associated signaling, 1 µM A_2AR selective antagonist ZM241385, 10 µM PKA inhibitor H-89, 5 µM PKC inhibitor GF109203X, and mouse mGluR5 siRNA (Santa Cruz Biotechnology, Inc.) and 1 µM mGluR5 antagonist MPEP (Pacheco et al., 2006) were used. For these cotreatments, CGS21680, ZM241385, and MPEP were added 10 min before LPS treatment. H-89 or GF109203X was added to the cultures 30 min before glutamate was added (Dai et al., 2010b). For the experiments involving mGluR5 siRNA, mouse neutrophils were transfected using the Nucleofection technology (Amaxa) for 24 h: 1.4 µg of each siRNA was transfected into 1 × 10⁶ cells according to the manufacturer’s recommendations. Controls underwent the same transfection procedure. Transfection efficiency was confirmed by Western blot for mGluR5 expression.

6 h after agents were added, total RNA of neutrophils was isolated using TRIzol reverse transcribed. IL-1β mRNA expression in neutrophils was evaluated with real-time PCR (Takara Bio Inc.) using the primers described previously (Dai et al., 2010b; Keita et al., 2010). All reported values are presented as mean ± SEM of three independent experiments conducted in triplicate.

By using protein A–agarose beads (EMD Millipore) and with the anti-A_2AR (Santa Cruz Biotechnology, Inc.), anti-mGluR2/3 (Abcam), or anti-mGluR5 antibodies (Abcam) for human and mouse, co-IP was performed as described previously (Schröder et al., 2009). The immunoreactive bands were developed with the enhanced chemiluminescence detection kit (SuperSignal; Thermo Fisher Scientific).

To investigate the associated signaling of A_2AR activation, Western blot analysis was also performed using antibodies against AC (Abcam) and PLC (Santa Cruz Biotechnology, Inc.). In these experiments, β-actin (Santa Cruz Biotechnology, Inc.) served as the endogenous control.

Statistical comparisons of more than two groups were performed by factorial ANOVA followed by Bonferroni’s post hoc test. Graphic data represent mean ± SEM. A value of P < 0.01 was considered statistically significant.

Table S1 lists characteristics of patients with severe TBI-induced ALI, including age, initial Glasgow coma scale, and the time of ALI onset. Table S2 lists characteristics of patients with lung trauma–induced ALI.

We gratefully acknowledge Dr. Susan E. Lewis for critically reading and editing the manuscript. We also thank Dr. Ping Li, Dr. Xing-Yun Chen, Mr. Ren-Ping Xiong, Mr. Zi-Ai Zhao, and Xing Chen for technical assistance.

This work was supported by grants from the National Natural Science Foundation of China (nos. 30900587, 81172817, and 31171022), the Natural Science Foundation of Chongqing, China (no. CSTC2009BB5317), and the Science Foundation of Third Military Medical University (to S.-S. Dai).

The authors have no conflicting financial interests.

Author contributions: S.-S. Dai and Y.-G. Zhou designed the research; S.-S. Dai, H. Wang, N. Yang, J.-H. An, W. Li, and Y.-L. Ning performed the research; Y.-G. Zhou and P.-F. Zhu contributed analytical tools; S.-S. Dai, J.-F. Chen, and Y.-G. Zhou analyzed the data; and S.-S. Dai and Y.-G. Zhou wrote the paper.