Promoted Interaction of Nuclear Factor-κB With Demethylated Purinergic P2X3 Receptor Gene Contributes to Neuropathic Pain in Rats With Diabetes

The lifetime incidence of diabetic neuropathy is 54–59% for patients with type 1 diabetes and 45–70% for patients with type 2 diabetes (1,2). The most common form of diabetic neuropathy is nerve damage in the periphery. Painful diabetic neuropathy develops in about a third of patients with diabetic peripheral neuropathy, who experience aberrant pain sensation, including spontaneous pain, hyperalgesia, and allodynia (3,4). Although several analgesic drugs, such as duloxetine and pregabalin, recently have been used in clinical practice, the pharmacological treatment of chronic painful diabetic neuropathy remains a challenge for the physician, partly because of our poor understanding of the molecular mechanisms underlying diabetic neuropathic pain (5,6).

Recent studies (7–9) have suggested the involvement of some ion channels or receptors in dorsal root ganglion (DRG) neurons in diabetic neuropathic pain in rats. Voltage-gated sodium channel Na_V1.7 expression was significantly increased in L4–L6 DRG neurons of rats with diabetic neuropathic pain (7,10). Transient receptor potential vanilloid 1 receptors are also thought to contribute to changes in the activity of DRG neurons under diabetic conditions (11). More recently, others and we have reported that the purinergic P2X receptor (P2XR) antagonists reduced pain behaviors in diabetic mice (12) and rats (13). ATP, an endogenous ligand for purinergic P2XRs, is present in abundance within neuronal and glial cells (14). It is released upon cell stress, stimulation, or damage, thus activating P2XRs present in sensory afferent endings to produce pain (15). Purinergic P2X ligand-gated ion channel 3 receptors (P2X3Rs) and P2X2Rs/P2X3Rs are preferentially expressed in DRG neurons and are upregulated under neuropathic (16–18), inflammatory (19), and visceral pain hypersensitivity conditions (20). However, the underlying mechanism by which p2xr gene expression is upregulated remains unknown under diabetic conditions.

Nuclear factor-κB (NF-κB), a nuclear transcription factor, is considered to control numerous genes encoding inflammatory and nociceptive mediators and to play roles in the development of thermal hyperalgesia after chronic compression of the DRG in rats (21,22). Our previous work (9,23) has demonstrated that p65 (a subunit of NF-κB) expression was upregulated in the DRGs of diabetic rats that are in pain and that the inhibition of p65 subunits by pyrrolidine dithiocarbamate (PDTC) (an inhibitor of NF-κB that inhibits the inhibitor of κB [IκB] phosphorylation) significantly alleviated mechanical allodynia of the rat hindpaw. However, whether p65 regulated P2X3R function and expression remains unclear. Therefore, we hypothesized that streptozocin (STZ)–induced diabetes led to demethylation of p2x3r gene promoter DNA and promoted p65 binding to the demethylated p2x3r gene promoter, thus resulting in the upregulation of P2X3R expression and diabetic pain hypersensitivity. To test this hypothesis, we constructed a lentiviral vector (LV) encoding a specially designed small interfering RNA (siRNA) against p65 gene (LV-p65) and observed the inhibitory effects of in vivo intrathecal delivery of LV siRNA. Our data demonstrate for the first time, to the best of our knowledge, that STZ-induced diabetes promoted p65 binding to the p2x3r gene promoter region. Selective blockade of p65 by PDTC or siRNA attenuates diabetes-induced painful behaviors via the inhibition of P2X3R activities of lumber DRG neurons that innervate the hindpaws. Therefore, selective inhibition of p65 represents a novel potential target for diabetic neuropathic pain therapy.

All experiments were approved by the Institutional Animal Care and Use Committee at Soochow University. Adult female Sprague-Dawley (SD) rats weighing 150–180 g were housed four per cage under a 12-h light-dark cycle and in a temperature-controlled room (25 ± 1°C). Rats were allowed access to tap water and standard laboratory chow ad libitum. The diabetic model was induced by a single intraperitoneal injection of STZ (65 mg/kg; Sigma-Aldrich, St. Louis, MO), as described previously (9,13). The average blood glucose levels are shown in Supplementary Table 1. Only rats with a blood glucose concentration of >15.0 mmol/L (270 mg/dL) were further used in the study.

All animals were habituated to the test environment for 1 week prior to testing. The pain threshold was determined before and after an injection of STZ or normal saline (NS). Rat paw withdrawal threshold (PWT) in response to the von Frey filament stimulation and thermal paw withdrawal latency (PWL) in response to radiant heat application to the plantar surface of the hindpaw were determined as described previously (13,23–25). All behavioral studies were performed under blind conditions.

The origin of the primary afferent innervation of the hindpaw was determined by retrograde tracing using 1,19-dioleyl-3,3,39,3-tetramethylindocarbocyanine methanesulfonate (DiI; Invitrogen, Carlsbad, CA), as described previously (13,26).

Isolation of DRG neurons from adult SD rats was described previously (9,19). DiI-labeled neurons were identified under an inverting fluorescence microscope (model IX71; Olympus, Tokyo, Japan). ATP-evoked currents were recorded by whole-cell patch-clamp recordings in a voltage-clamped mode, as described previously (13).

The expressions of P2XRs (P2X1, P2X2, and P2X3) and p65 in L4–L6 DRGs from control and diabetic rats with or without insulin treatment were measured using Western blot analyses (9,13). Anti-p65 (1:500; Abcam, Hong Kong) and anti-P2XRs (1:1,000; Abcam, Cambridge, MA) and corresponding horseradish peroxidase–conjugated secondary anti-rat antibodies at dilutions of 1:2,000 and 1:4,000 were used to probe the proteins.

The mRNA expressions of p65, P2X3R, and DNA methylation-related enzymes were measured using real-time PCR (RT-PCR), as described previously (9,27). Total RNA was extracted from L4–L6 DRGs with TRIzol (Invitrogen), and cDNA was synthesized from total RNA using an Omniscript RT kit 150 (QIAGEN) following the instructions of the supplier. The sequences of the specific primers are shown in Table 1.

Triple labeling was performed as described previously (9,28). DiI was injected into the plantar skin of bilateral hindpaws to label hindpaws innervating DRG neurons. p65 (1:100) and P2X3R (1:200) antibodies were used in the current study.

The methylation status of the cytosine guanine dinucleotide (CpG) islands of all samples was initially screened at p2x3r gene promoter regions by methylation-specific PCR (MSP), as described previously (27). The MSP primers designed for P2X3R are shown in Table 1. The methylation status was further validated by bisulfite sequencing, as described previously (27,29). Specific primers were designed to amplify CpG-rich regions in the p2x3r promoter (Table 1).

A chromatin immunoprecipitation (ChIP) assay was performed using the Upstate Kit (EMD Millipore, Billerica, MA) according to the manufacturer's instructions, as described previously (9,30). Five NF-κB consensus site–specific primers in the p2x3r promoter are shown in Table 1.

As described previously (31), the siRNA targeting the cDNA sequence of rat p65 (GenBank Accession #NM_199267) was used. The sequences of the siRNA were as follows: 5-GCAGUUCGAUGCUGAUGAAUU-3. An additional scrambled sequence was also designed as a negative control (NC). The replication-deficient, self-inactivating LV pFU-GW-RNAi-GFP (Shanghai Genechem Co., Ltd.) was generated. The cDNAs corresponding to the p65 siRNA and NC were subcloned into the vector pFU-GW-RNAi-GFP. The resulting recombinant LVs were designated as LV-p65 siRNA or LV-NC. The titer of lentivirus was determined by a hole-by-dilution titer assay. The final titer of LV-p65 siRNA and LV-NC was 1 × 10⁹ transduction units (TU)/mL. To determine the level of efficiency of infection of virus constructs, a pilot study with LV-GFP was performed. The lentiviruses were then injected into rats intrathecally. Experiments were performed 4 weeks later.

Data were expressed as the mean ± SEM. Software OriginPro 8 (OriginLab, Northampton, MA), MATLAB (MathWorks, Natick, MA), and SPSS version 17.0 were used for data analysis. Significance was determined using a two-sample t test, Mann-Whitney U test, Dunn post hoc test following Friedman ANOVA, two-way repeated-measures ANOVA followed by Tukey post hoc test, or χ² test, as appropriate. A P value <0.05 was considered to be statistically significant.

Consistent with our previous studies (13,23), a single injection of STZ produced hindpaw allodynia in rats (Fig. 1A, n = 10 rats for each group, **P < 0.01 compared with control rats, Mann-Whitney U test; Fig. 1B, n = 10 rats for each group, **P < 0.01, compared with control rats, two-sample t test). Intrathecal injection of suramin, a nonselective P2 receptor antagonist, attenuated the PWT in a dose-dependent and time-dependent manner in diabetic rats (Fig. 1C, n = 6 rats for each group, **P < 0.01, compared with NS, Friedman ANOVA). A similar effect was found in PWL (Fig. 1D, n = 6 rats for each group, *P < 0.05, **P < 0.01, compared with NS, two-way repeated-measures ANOVA followed by Tukey post hoc test). The maximal inhibitory effects for suramin were observed at a dose of 500 μg/kg. Intrathecal injection of A-317491, a potent antagonist for P2X3Rs and P2X2Rs/P2X3Rs, significantly attenuated the PWT (Fig. 1E, n = 6 rats for each group, *P < 0.05, compared with NS, Friedman ANOVA) and PWL (Fig. 1F, n = 6 rats for each group, **P < 0.01, compared with NS, two-way repeated-measures ANOVA followed by Tukey post hoc test) in diabetic rats. The maximal inhibitory effects were observed at a dose of 300 nmol/kg for A-317491. Of note is that suramin at a dose of 50 μg/kg or A-317491 at a dose of 300 nmol/kg did not produce any effect on PWT and PWL in healthy control rats (Fig. 1G and H, n = 6 rats for each group).

To test the hypothesis that ATP signaling is enhanced in diabetic rats, currents evoked by ATP in acutely isolated DRG neurons were measured using whole-cell patch-clamp recording techniques. Hindpaw-innervating DRG neurons were labeled by the fluorescent dye DiI (Fig. 2A, top, arrow). At a holding potential of −60 mV, ATP (20 μmol/L) evoked the following two types of inward currents, as described previously (19): fast inactivating currents and slow inactivating currents. Since the slow inactivating currents was evoked in <10% of neurons and this figure did not change in STZ-induced diabetes, only the fast inactivating currents were included in the current study (Fig. 2B, top). The average peak current density obtained from diabetic rats was significantly increased (Fig. 2B, bottom, n = 11 cells for each group, **P < 0.01 compared with control rats, two-sample t test). In addition, the expression of both P2X3R mRNA and proteins was significantly higher in diabetes than in control rats (Fig. 2C and D, n = 4 rats for each group, *P < 0.05, **P < 0.01, compared with control rats, two-sample t test). In contrast, the expression of P2X1 and P2X2 receptors was not significantly altered 4 weeks after STZ injection (Fig. 2E and F, n = 6 for control rats, n = 3 for diabetic rats).

To determine whether an epigenetic mechanism contributes to the upregulation of P2X3R expression, DNA methylation status at the promoter region of the p2x3r gene was examined with MSP and a bisulfite genomic sequence assay. The genomic structure of rat p2x3r gene contains one CpG island in the promoter region (Fig. 3A). When the frequency of CpG is high in certain sections of the genome, these segments are known as CpG islands. Methylation occurs only in the CpG dinucleotide sites. All samples yielded methylated and unmethylated bands from control and diabetic rats, as determined by MSP assay (Fig. 3B). The ratio of methylated to unmethylated CpG sites was greatly reduced (Fig. 3B, n = 8 for each group, *P < 0.05, compared with control rats, two-sample t test). To further confirm the methylation status of CpG sites within the p2x3r promoter, DNA sequencing was performed on PCR products of the 406-bp fragment (−1,737 to −1,332) obtained after the treatment of genomic DNA samples with sodium bisulfite. As shown in Fig. 3A, the 406-bp fragment contains 18 CpG sites. All DRG samples from control and diabetic rats were successfully sequenced (Fig. 3C). The degree of methylation of the CpG sites within the fragment was significantly reduced in DRGs 4 weeks after the induction of diabetes (Fig. 3D, n = 4 for each group, *P < 0.05, compared with control rats, χ² test), which is consistent with the results of the MSP assay. In addition, the expression level of DNA methyltransferase (DNMT) 3b was remarkably reduced, while the expression level of DNMT3a was not altered in diabetic rats (Fig. 3E, n = 4 for each group, *P < 0.05, compared with control rats, two-sample t test). We also examined the expression of growth arrest and DNA damage–inducible protein α (Gadd45a), methyl-binding domain protein (MBD) 2, MBD4, and thymine DNA glycosylase (TDG) of L4–L6 DRGs in control and diabetic rats. None of these molecules was altered in their expression in diabetic rats (Fig. 3F, n = 4 for each group).

We next determined whether NF-κB signaling was involved in P2X3R expression. The p65 antibody used in the study detects the total expression of p65, including phosphorylated and total proteins, and it was not a phosphorylation-independent antibody. p65 was coexpressed in P2X3Rs-positive hindpaw-innervating DRG neurons (Fig. 4A). As indicated in Fig. 4B, five transcription-binding sites for p65 were found in the p2x3r gene promoter (http://www.gene-regulation.com/pub/programs/alibaba2/index.html). The ChIP assay demonstrated a dramatic increase in p65 binding activities to the p2x3r promoter region for p65 binding site 1 (Fig. 4C, n = 4 for each group, *P < 0.05, compared with control rats, two-sample t test). p65 binding sites 4 (Fig. 4D, n = 4 for each group) and 5 (Fig. 4E, n = 4 for each group) in diabetic rats were not remarkably altered when compared with age- and sex-matched control rats. The predicted p65 binding sites 2 and 3 were not confirmed by the ChIP assay in this study (data not shown). Nevertheless, these data indicated an increase in the interaction between p65 and the p2x3r promoter region in STZ-induced diabetes.

To determine whether NF-κB plays a role in the development of diabetic pain hypersensitivity, we examined the expression of p65 in L4–L6 DRGs in STZ-induced diabetes. The expression of p65 both at the protein level (Fig. 5A, n = 3 for control rats, n = 4 for diabetic rats, *P < 0.05, compared with control rats, two-sample t test) and the mRNA level (Fig. 5B, n = 4 for control rats, n = 8 for diabetic rats, *P < 0.05, compared with control rats, two-sample t test) was significantly enhanced in L4–L6 DRGs 4 weeks after STZ injection when compared with age- and sex-matched control rats. Intraperitoneal injection of PDTC (50 mg/kg, once daily for 7 consecutive days starting 3 weeks after the STZ injection) greatly attenuated the PWT and PWL (Fig. 5C and D, n = 8 rats for each group, **P < 0.01, compared with the NS group, two-sample t test). PDTC (50 mg/kg) did not produce any effect on PWT and PWL in control rats (Fig. 5E and F, n = 5 for each group). PDTC treatment also markedly suppressed P2X3R expression at the protein level (Fig. 5G, n = 6 rats for each group, **P < 0.01, compared with the NS group, two-sample t test) and mRNA level in diabetic rats (Fig. 5H, n = 6 rats for NS, n = 3 rats for PDTC, *P < 0.05, compared with the NS group, Mann-Whitney U test). The results suggested that p65 plays a role in diabetic pain hypersensitivity through the regulation of P2X3R expression.

To further confirm the role of p65 signaling in diabetic pain hypersensitivity, we applied p65 siRNA to specifically knock down p65 expression. We first developed a highly efficient method of lentivirus-mediated delivery of siRNA targeting of p65 for gene silencing, as described previously (31). This method successfully transduced LV-GFP into cultured DRG neurons in vitro (data not shown) and in vivo by a single intrathecal injection. As shown in Fig. 6, hindpaw-innervating L5 DRG neurons were labeled by DiI (Fig. 6A and D, red, arrows). Green signals indicated that the LV-GFP-NC siRNA (Fig. 6B) or LV-GFP-p65 siRNA (Fig. 6E) was successfully transduced into these neurons, respectively. P2X3R-positive neurons were labeled in blue in Fig. 6C and F. The arrows in Fig. 6B and E indicate that these neurons were DiI and GFP double positive. The empty arrowheads in Fig. 6F indicated that these neurons were DiI and GFP double positive, but P2X3R negative.

After intrathecal injection of LV-p65 siRNA (1 × 10⁷ TU) in diabetic rats, the percentage of P2X3R-positive neurons in DiI-labeled neurons was significantly decreased (Fig. 6G, n = 4 for the each group, **P < 0.01, compared with NC siRNA, two-sample t test). The expression of p65 was also remarkably decreased in L4–L6 DRGs (Fig. 6H, n = 4 for each group, *P < 0.05, compared with NC siRNA, two-sample t test). Importantly, the delivery of siRNA targeting p65 for gene silencing also decreased the expression of P2X3Rs (Fig. 6I, n = 4 for each group, *P < 0.05, compared with NC siRNA, two-sample t test).

In addition, LV-p65 siRNA treatment markedly suppressed the current evoked by ATP in diabetic rats (Fig. 7A and B, n = 12 cells for NC siRNA, n = 11 cells for p65 siRNA, **P < 0.01, compared with NC siRNA, two-sample t test). The delivery of siRNA targeting p65 significantly attenuated mechanical allodynia (Fig. 7C, n = 8 for each group, **P < 0.01, compared with NC siRNA, Friedman ANOVA) and thermal hyperalgesia (Fig. 7D, n = 8 for each group, **P < 0.01, compared with NC siR group, Friedman ANOVA). However, LV-p65 siRNA did not produce any effect on PWT and PWL in control rats (Fig. 7E and F, n = 5 rats for each group). These results further suggested that p65 might regulate the function of P2X3Rs through regulating its expression, thus contributing to diabetic pain hypersensitivity.

We next determined whether insulin treatment attenuated mechanical allodynia and thermal hyperalgesia and reversed changes in the expression of p65 and P2X3Rs in diabetic rats. Starting on day 8 after STZ injection, each rat was given two daily subcutaneous injections of either 1–4 units of insulin (human recombinant insulin; Novo Nordisk) in 0.2 mL or the same volume of NS as a control for 21 days. The insulin dose was adjusted to correct hyperglycemia and to avoid hypoglycemia as well (32). The average blood glucose level was significantly decreased 1 week later and was maintained at a relative low level for another 2 weeks within our observation period of time, and the body weight of insulin-treated rats exhibited a steady increase (Supplementary Table 2; n = 7 for each group, **P < 0.01, compared with NS, two-way repeated-measures ANOVA followed by Tukey post hoc test). In addition, insulin treatment reversed PWT and PWL when compared with NS (Fig. 8A and B, n = 7 for each group, *P < 0.05, **P < 0.01, compared with NS, Friedman ANOVA). This inhibitory effect was exhibited from 1 to 3 weeks after insulin treatment. More importantly, insulin treatment significantly downregulated the expressions of p65 and P2X3Rs in L4–L6 DRGs in diabetic rats (Fig. 8C and D, n = 4 for each group, *P < 0.05, **P < 0.01, compared with NS, two-sample t test). These data suggest that insulin loss or/and hyperglycemia but not STZ neuronal toxicity (33) are most likely the causes for many of the biochemical disorders observed in the current study.

We demonstrated that P2X3Rs play an important role in the development of peripheral diabetic neuropathic pain. In our previous study (13), we have reported the role of P2X3Rs in diabetic allodynia 2 weeks after STZ injection. In the current study, we further extended our observation time period to 4 weeks after STZ injection since STZ-induced mechanical allodynia and thermal hyperalgesia were attenuated by a nonselective P2 receptor inhibitor suramin and by a potent P2X3R and P2X2R/P2X3R antagonist, A-317491. In addition, we showed that P2X3R function was greatly enhanced, which was likely mediated by the upregulation of P2X3R expression 4 weeks after STZ induction of diabetes. This result is different from our previous report that membrane P2X3R protein expression was significantly upregulated without alteration in the expression of total P2X3R proteins in rat DRGs 2 weeks after STZ injection (13). Although the detailed mechanisms are unknown, these results suggest that mechanisms underlying the potentiation of ATP-induced responses are different with the progress of the diabetes. At the beginning of diabetes (i.e., 2 weeks), the enhanced ATP-induced responses might be due to the promoted trafficking of P2X3Rs from cytoplasm to cell membrane (13). With the development of diabetic mechanical allodynia and thermal hyperalgesia (i.e., 4 weeks), the upregulation of P2X3R expression might be a major contributor to the potentiation of ATP-induced responses. However, the upregulation is gene specific since the expression of P2X1R and P2X2R was not significantly altered in this setting. Therefore, it is most likely the case that the p2x3r gene might be an effective facilitator of STZ-induced diabetic pain hypersensitivity. Once activated, P2X3Rs cause pain, most likely because of increased drive from the periphery, increased spontaneous activity of DRG neurons, and/or increased sensitivity in the dorsal horn of the spinal cord (34,35).

The most important finding is that p2x3r gene was significantly demethylated in diabetic rats. Emerging evidence (36,37) shows that epigenetic regulations (i.e., aberrant DNA methylation and histone acetylation) are involved in gene expression. Gene promoter DNA methylation suppresses gene expression, whereas gene promoter DNA demethylation facilitates gene upregulation (38,39). Wang et al. (40) reported a role of DNA methylation in nerve injury–induced pain. We have previously reported that DNA demethylation of the cbs gene promoter is involved in inflammatory hyperalgesia (27) and gastric hypersensitivity (9), which is consistent with the upregulation of cystathionine β-synthetase (an endogenous synthetase of hydrogen sulfide) and mRNA. Methylation occurs only in the CpG dinucleotide sites (41,42). The increase in demethylation of the p2x3r promoter region would give rise to an upregulation of P2X3R expression in diabetic rats. To date, the mechanism of DNA methylation remains largely unknown. In mammals, there are four types of DNMTs in regulating methylation status (38). DNMT3a and DNMT3b, two major enzymes for de novo methyltransferase, are mainly responsible for the methylation of CpG islands (43,44). The reduction of the methylase (i.e., DNMT3b) can lead to the demethylation state, although the expression of demethylation enzymes such as TDG, MBD2, MBD4, and Gadd45a was not significantly affected in STZ-induced diabetes. This is consistent with our previous report (9) that DNMT3a and DNMT3b were drastically downregulated in gastric DRGs under diabetic pain hypersensitivity conditions. This is, however, different from the finding in our previously published study (27) that an active mechanism is involved in inflammatory pain. Although the detailed mechanisms for this difference are unknown, it is most likely that the mechanism of diabetes-induced DNA demethylation is different from that of inflammation-induced DNA methylation but may share a common mechanism for neuropathic pain.

Demethylation of a gene promoter region leads to the exposure of a transcription factor binding site, thus allowing the transcription factor to bind. p65 plays an important role in inflammatory diseases, immune responses, and cell death by binding to the promoter region of various genes such as tumor necrosis factor-α (TNF-α), interleukin-1β, cyclooxygenase-2, and inducible nitric oxide synthase (45,46). We demonstrated that diabetes promoted the binding ability of p65 with demethylation of the p2x3r gene. In the development of diabetic neuropathy, p65 is the central link of multiple pathways. It can be activated by reactive oxygen species, advanced glycation end products, and other cytokines such as TNF-α (47,48). The activation of NF-κB can upregulate P2X3R expression and promote the production of cytokines such as TNF-α (45,46), thus forming a positive feedback loop to maintain the diabetic pain hypersensitivity. Of note is that Berti-Mattera et al. (49) showed a reduction in p65 in nuclear fractions of DRG in diabetic rats. Although the reasons for the difference are not clear, the disease duration (3 months vs. 4 weeks), the animal species (Lewis rats vs. SD rats), and the dosage of STZ used (60 vs. 65 mg/kg body wt) could be the causes. Since insulin treatment significantly attenuated the pain hypersensitivity in an association with the reduced expression of p65 and P2X3Rs, it is likely that insulin loss and/or hyperglycemia might be the triggers of gene expression. Future studies need to investigate the mechanisms in detail and the possible role of NF-κB p50 subunits in p65-mediated gene expression.

In conclusion, our findings demonstrated for the first time, to the best of our knowledge, that the upregulation of P2X3R expression is attributed to p2x3r gene promoter DNA demethylation and p65 activation. The enhanced interaction of p65 with the p2x3r gene promoter would contribute to abnormal diabetic peripheral neuropathic pain. The results may provide promising clues for the development of new therapeutic strategies for managing intractable neuropathic pain in patients with diabetes.

Funding. This work was supported by grants from National Natural Science Foundation of China (31400947 to H.-H.Z., 81471041 to J.H., 31271258 to X.J., and 81230024 and 81471137 to G.-Y.X.) and from the Priority Academic Program Development of Jiangsu Higher Education Institutions of China.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. H.-H.Z. and J.H. researched and analyzed the data and wrote the article. Y.-L.Z. researched and analyzed the data. X.Q., Z.-Y.S., P.-P.Y., and S.H. researched the data. X.J. contributed to the discussion and reviewed and edited the article. G.-Y.X. designed and supervised the experiments and edited the article. G.-Y.X. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.