Opposing roles for SNAP23 in secretion in exocrine and endocrine pancreatic cells

Newly synthesized secretory proteins, such as hormones and digestive enzymes, are secreted through the fusion of secretory vesicles with the plasma membrane, which is mediated by SNARE proteins (Hong, 2005; Jahn and Scheller, 2006; Südhof and Rothman, 2009; Thorn and Gaisano, 2012; Hong and Lev, 2014). One v-SNARE protein, namely vesicle-associated membrane protein (VAMP), and two t-SNARE proteins, namely syntaxin and SNAP, form an α-helical ternary complex to initiate membrane fusion. Regulation of exocytosis by SNAREs is essential for the secretion of neurotransmitters, hormones, enzymes, and cytokines from various tissues and cells.

In humans, 38 SNARE proteins have been identified (Hong and Lev, 2014). Each SNARE protein exhibits distinct tissue expression and intracellular localization as well as complex formation with its appropriate partners. For example, the VAMP2–syntaxin1A–SNAP25 complex, which is the best-characterized neuronal SNARE complex, catalyzes synaptic vesicle fusion at the presynaptic terminal (Sutton et al., 1998). This complex also promotes the fusion of insulin granules in pancreatic β cells (Hou et al., 2009; Kasai et al., 2012).

SNAP23 is a ubiquitously expressed homologue of SNAP25 (Ravichandran et al., 1996; Wang et al., 1997). Although SNAP25 plays critical roles in neurotransmitter release from neurons and insulin secretion from pancreatic β cells (Washbourne et al., 2002; Takahashi et al., 2010), SNAP23 is involved in exocytotic events in diverse nonneuronal cells, such as surfactant release from alveolar epithelial cells, glucose transporter GLUT4 translocation in adipocytes, and Ig release from plasma cells (Kawanishi et al., 2000; Abonyo et al., 2004; Reales et al., 2005). In pancreatic β cells, SNAP23 also promotes the fusion of insulin granules to the plasma membrane (Sadoul et al., 1997), whereas in pancreatic acinar cells, SNAP23 binds VAMP2 or VAMP8 for the fusion of amylase granules to the plasma membrane (Wang et al., 2004; Weng et al., 2007; Cosen-Binker et al., 2008). Thus, in vitro studies have suggested that SNAP23 is physiologically essential. However, the in vivo function of SNAP23 in the fusion of secretory granules remains largely unknown.

In this study, we generated pancreatic exocrine- or endocrine-specific Snap23 knockout (KO) mice to investigate the in vivo function of SNAP23. The exocrine-specific KO mice showed decreased fusion of zymogen granules (ZGs), but the endocrine-specific KO mice showed increased fusion of insulin granules and improved glucose tolerance. These results suggest that SNAP23 plays opposite roles in secretion in the exocrine and endocrine pancreas. Furthermore, we found that the SNAP23-binding compound MF286 promoted insulin secretion and improved glucose tolerance by inhibiting formation of the SNARE complex that includes SNAP23. As MF286 also inhibits amylase secretion from the exocrine pancreas, as seen in exocrine-specific Snap23 KO mice, our study indicated that MF286 might be a candidate drug for diabetes and pancreatitis treatment.

To determine the in vivo function of SNAP23, we generated Snap23 conditional KO mice using a revertible KO system (Sato et al., 2007; Fig. 1 A). Consistent with a previous study (Suh et al., 2011), the homozygous mutant mice (Snap23^−/− and Snap23^geo/geo) exhibited embryonic lethality before 8.5 d postcoitum (Fig. 1 B).

In the mouse and human pancreas, SNAP23 was expressed in both the pancreatic islets (endocrine) and acini (exocrine), but SNAP25 was expressed only in the islets (Fig. 1, C and D). Because these data suggest that SNAP23 participates in both the secretion of insulin from β cells and the secretion of digestive enzymes from acinar cells, we generated acinar cell (exocrine)–specific KO (AcKO) mice and β cell (endocrine)–specific KO (BcKO) mice. The AcKO mice (Elastase-Cre; Snap23^floxed/− or ^{floxed/floxed}) were generated by crossing Snap23 floxed mice (Snap23^{floxed/floxed}) with Elastase-Cre mice (Hashimoto et al., 2008) expressing Cre recombinase by the elastase promoter. BcKO mice (rat insulin promoter II [RIP]–Cre; Snap23^floxed/− or ^{floxed/floxed}) were generated by crossing Snap23^{floxed/floxed} with RIP-Cre mice expressing Cre recombinase by RIP (Herrera, 2000; Kitamura et al., 2009). Pancreatic and duodenal homeobox gene [Pdx] 1–Cre-derived conditional KO (PcKO; Pdx1-Cre; Snap23^floxed/− or ^{floxed/floxed}) mice were generated by crossing with Pdx1-Cre mice expressing Cre recombinase by the Pdx1 promoter (Gu et al., 2002). AcKO, BcKO, and PcKO mice were born at the expected Mendelian frequency and appeared overtly normal.

The AcKO mice, namely the pancreatic acinar cell (exocrine)–specific SNAP23 KO mice, appeared healthy. The body weights and serum biochemical figures were similar between control and AcKO mice (Fig. 2 A and Table 1). Immunofluorescence and Western blot analysis showed that SNAP23 levels on the plasma membrane were greatly decreased in acinar cells (Fig. 2, B and C).

The v-SNARE proteins VAMP2 and VAMP8 are crucial for the exocytosis of ZGs in the acinar cells, and SNAP23 is suggested to be a binding partner of VAMP2 or VAMP8 (Wang et al., 2004; Weng et al., 2007; Cosen-Binker et al., 2008). Therefore, we hypothesized that SNAP23 is involved in ZG exocytosis. To test this, we analyzed the density of ZGs in the acinar cells by light and electron microscopy. First, we observed the morphology and size of acinar cells by hematoxylin and eosin (HE) staining, but we could find no difference between control and AcKO acinar cells (Fig. 2 D). We next used transmission electron microscopy (TEM), but, again, the ZG density in AcKO cells was not significantly different. However, after cholecystokinin (CCK) stimulation, which evokes secretion of digestive enzymes by fusion of the ZGs with the apical membrane, the ZG density in AcKO cells was significantly higher (Fig. 2, E and F). This result suggested that ZG exocytosis was reduced in AcKO acinar cells.

To determine whether the secretion of digestive enzyme was reduced in AcKO acinar cells, we performed an in vitro amylase secretion assay using a colorimetric assay. Consistent with the result from the TEM analysis, we confirmed the reduction of amylase secretion in AcKO cells after CCK stimulation (Fig. 2 G). Furthermore, to check the effects of altered lipase secretion, we examined the lipids in stools. In AcKO stools, Sudan III staining showed many lipid droplets (Fig. 2 H), and percentage of fecal triglyceride was significantly increased (Fig. 2 I). These results suggest that the decreased lipase secretion leads to the lipid indigestion in AcKO mice.

ZGs are secreted by compound exocytosis, which is caused by an increase in intracellular Ca²⁺ concentration. In this process, primary fusion events occur in which the first granules fuse with the cell membrane, and this is followed by sequential fusion events in which secondary granules fuse onto these primary granules (Nemoto et al., 2001; Pickett and Edwardson, 2006; Kasai et al., 2012; Thorn and Gaisano, 2012). To investigate the frequency of ZG fusion, we counted the number of exocytotic events and monitored intracellular Ca²⁺ signaling using two-photon microscopy (Nemoto et al., 2001). Isolated acinar cells were immersed in a solution containing sulforhodamine B (SRB) as a fluid-phase tracer, and the primary and sequential exocytotic events were detected as the abrupt appearance of small fluorescent spots (Fig. 3 and Video 1). In the AcKO cells, both the primary and sequential exocytotic fusion events were significantly reduced (Fig. 3, A–E). The ratio of sequential exocytotic events to primary exocytotic events was lower in the AcKO cells, suggesting that SNAP23 participates in both primary and sequential secretion (Fig. 3 F). In contrast to the difference in the exocytotic events, the intracellular Ca²⁺ oscillations and maximal increase during CCK stimulation were similar in the control and AcKO cells, excluding that the inhibition of fusion events was a result of diminished Ca²⁺ concentration (Fig. 3, G and H). Collectively, we concluded that SNAP23 is essential for compound exocytosis in pancreatic acinar cells.

SNAP23 is also expressed in other exocrine tissues such as salivary glands (Wang et al., 2007). To confirm whether SNAP23 participates in the secretion in exocrine system in general, we measured the amylase secretion from parotid exocrine cells. Parotid exocrine cells were isolated from Snap23 floxed mice (Snap23^{floxed/floxed}) and cultured with a Cre recombinase–encoding adenovirus (Ad-Cre) or with a control adenovirus (Ad-LacZ; Fig. 4 A). 48 h after infection, mRNA level of SNAP23 was greatly decreased (Fig. 4 B), and SNAP23 levels on the plasma membrane were decreased in Ad-Cre–treated cells (Fig. 4 C). Amylase secretion after isoproterenol (IPR) and methacholine (MeCh) stimulation was reduced in Ad-Cre–treated cells (Fig. 4 D). These results indicate that SNAP23 is also essential for secretion in other exocrine gland.

The BcKO mice (RIP-Cre; Snap23^{floxed/floxed} or ^floxed/−) also grew normally, and serum biochemical tests showed no differences compared with the control mice (Fig. 5 A and Table 1). By immunofluorescence, SNAP23 localized to the plasma membrane in the control β cells but was absent in the BcKO β cells, confirming the successful depletion of SNAP23 in the BcKO β cells (Fig. 5 B). In contrast, the staining patterns and intensities of other SNARE proteins, such as SNAP25, syntaxin1A, and VAMP2, were similar between the control and BcKO islets (Fig. 5 B). The deletion of SNAP23 did not cause any overt change in the morphology of the islets, confirmed by HE staining (Fig. 5 C). We also found no differences in the expression levels of insulin, glucagon, and somatostatin (Figs. S1 and S2).

A previous study reported that SNAP23 promotes insulin secretion in HIT cells (Sadoul et al., 1997). Therefore, we speculated that glucose tolerance would be impaired in the BcKO mice. The blood glucose levels in the fasting or free-feeding mice were similar between the control and BcKO mice. Surprisingly, however, the blood glucose levels in the fasting-refeeding BcKO mice were significantly lower (Fig. 6 A). Furthermore, the serum insulin levels in the fasting-refeeding BcKO mice increased more than twofold (Fig. 6 B).

To further investigate the role of SNAP23 in glucose tolerance, we performed an i.p. glucose tolerance test (IPGTT). In agreement with the fasting-refeeding experiments, glycemia in response to glucose stimulation was significantly reduced in the BcKO mice (Fig. 6 C). The amount of secreted insulin 15 min after glucose injection was also dramatically increased (Fig. 6 D). In contrast, an insulin tolerance test (ITT) showed that the insulin sensitivity in the peripheral tissue was similar (Fig. 6 E), demonstrating that the decline in blood glucose levels during IPGTT was the result of increased insulin secretion of BcKO β cells.

To obtain precise information about the kinetics of insulin exocytosis, we isolated the islets and examined the insulin secretion (Fig. 6, F–H). When the islets were incubated with a low concentration (2.2 mM) of glucose, BcKO islets secreted similar levels of insulin as control islets. However, upon stimulation with a high concentration (16.7 mM) of glucose, BcKO islets secreted a significantly higher amount of insulin (Fig. 6 F).

There are at least two phases of the insulin secretion process: the initial rapid first phase and the sustained second phase (Hou et al., 2009). To check this secretion process, we performed a perfusion analysis in the isolated islets. The amount of secreted insulin was increased only during the first phase in the BcKO-perfused islets (Fig. 6, G and H). Additionally, we expressed insulin-GFP in β cells and observed the exocytotic events using total internal reflection fluorescence microscopy (TIRFM). The experiment revealed that the fusion events of the predocked granules but not the newcomer granules were increased in the BcKO islets (Fig. 6, I and J). These results suggest that SNAP23 inhibits the first phase of secretion by suppressing the fusion of predocked granules.

To confirm the phenotypes of BcKO mice, we generated additional SNAP23 PcKO mice (Gu et al., 2002). In the wild-type islets, SNAP23 was expressed in α and β cells but was scarcely expressed in δ cells, whereas SNAP25 was expressed in all three types of cells (Figs. S1 and S2). These data suggest that SNAP23 is involved in the secretion of insulin and glucagon. Because the Pdx1-Cre transgenic mice express the cre gene in all pancreatic cell types (Gu et al., 2002), we assumed that it recombined the Snap23 floxed allele in both α and β cells in the islets. Unexpectedly, our PcKO mice showed that SNAP23 was depleted in most of the β cells but was present in the α cells (Fig. S1). This phenotype might be caused by a difference in genetic background. It is also reported that a difference in the target floxed allele affects the recombination efficiency in a given cell (Zheng et al., 2000; Heffner et al., 2012).

Similar to BcKO mice, an IPGTT experiment demonstrated that glucose tolerance was improved in the PcKO mice (Fig. S3 A). Furthermore, the serum insulin levels 30 min after glucose stimulation were increased in the PcKO mice (Fig. S3 B). To observe insulin exocytosis from the PcKO β cells, we counted the number of fusion events in the β cells using two-photon microscopy (Takahashi et al., 2002) (Fig. S3 C and Video 2). Consistent with the TIRFM analysis of BcKO islets, the fusion of insulin granules occurred more frequently during the initial 5 min in the PcKO islets (Fig. S3, D and E). The intracellular Ca²⁺ concentrations were similar between the control and PcKO islets (Fig. S3 F), excluding the possibility that the increased fusion events were a result of increased Ca²⁺ concentration.

By TEM, the β cell morphology and granule sizes were indistinguishable between the control and PcKO islets (Fig. S3, G and H). In addition, the numbers of total and docked insulin granules were similar (Fig. S3, I and J). These results suggest that the increased insulin secretion was not caused by abnormalities in the insulin granule.

Among the SNARE proteins, VAMP2, syntaxin1A, and SNAP25 are involved in the fusion between the insulin granules and the plasma membrane (Regazzi et al., 1995; Sadoul et al., 1995; Ohara-Imaizumi et al., 2007; Takahashi et al., 2015). The SNAP25–VAMP2–syntaxin1A complex is reported to exhibit tighter binding among its components and can fuse membranes more efficiently than the SNAP23–VAMP2–syntaxin1A complex during exocytosis (Sørensen et al., 2003; Vites et al., 2008; Montana et al., 2009). Therefore, we speculated that SNAP23 might compete with SNAP25 for binding to syntaxin1A and VAMP2. To test this, we performed in vitro binding competition studies using recombinant SNARE proteins. We incubated GST-SNAP25 with equal amounts of His-VAMP2 and His-syntaxin1A in the presence of variable concentrations of His-SNAP23. As shown in Fig. 7 A and illustrated in Fig. 7 B, increasing SNAP23 concentration inhibited the formation of the SNAP25–syntaxin1A–VAMP2 complex, confirming that SNAP23 competes with SNAP25 for binding to syntaxin1A and VAMP2.

SNAP23 forms a homotetrameric complex using its N-terminal coiled-coil domain (Freedman et al., 2003). Because the amino acid sequence of the coiled-coil region of SNAP23 is similar to the same region of SNAP25, SNAP23 might bind SNAP25 and disturb the formation of the SNAP25-containing SNARE complex, which participates in insulin granule fusion. To exclude this possibility, we performed an immunoprecipitation assay using islet lysates from wild-type mice. The levels of SNAP23 that coimmunoprecipitated with the antibody against SNAP25 and control IgG were equally low in the islet lysate (Fig. 7 C). This result suggests that SNAP23 does not bind SNAP25 in the pancreatic islets.

In the BcKO islets, we found no differences in the levels of SNAP25, syntaxin1A, and VAMP2. However, immunoprecipitation assays of control and BcKO islets revealed that the disruption of SNAP23 increased the amount of SNAP25-bound VAMP2 (Fig. 7, D and E). This indicates that SNAP23 depletion allowed the increased formation of SNAP25–VAMP2–syntaxin1A complexes, further resulting in the enhanced insulin secretion in the BcKO islets.

To verify this result using a simplified system, we used the mouse insulinoma-derived cell line MIN6 as a model β cell. We first addressed whether the knockdown of SNAP23 in MIN6 cells facilitated hormone secretion. MIN6 cells were transfected with human growth hormone (hGH) together with control siRNA, siRNA against SNAP23, or siRNA against SNAP25. As expected, the hormone secretion was significantly enhanced in SNAP23-knockdown MIN6 cells, whereas it was decreased in SNAP25-knockdown MIN6 cells compared with control cells (Fig. 7 F). We next performed an immunoprecipitation assay similar to Fig. 7 D using the SNAP23-knockdown MIN6 cells. Consistent with the result obtained by BcKO islets, the amount of VAMP2 in SNAP25 immunoprecipitates increased in SNAP23-knockdown MIN6 cells (Fig. 7, G and H). Collectively, our results suggest that the increased hormone exocytosis in the BcKO islets and SNAP23-knockdown MIN6 cells was caused by the increased formation of SNAP25–VAMP2–syntaxin1A complexes as a consequence of SNAP23 depletion.

Because SNAP23 depletion caused increased insulin secretion, we speculated that a SNAP23 inhibitor might be a new therapeutic drug candidate for diabetes. We screened a chemical compound library from RIKEN's NPDepo (Zimmermann et al., 2013) with GST-SNAP23 and GST-SNAP25 and identified five compounds that bound SNAP23 more strongly than SNAP25. For further screening, the amounts of insulin secretion were measured after treatment of MIN6 cells with each compound. As a result, we found that after treatment with compound 2, MF286, the amount of insulin was significantly increased in response to glucose stimulation (Fig. 8, A and B). The influence of these compounds on MIN6 cell survival was low, which excluded the possibility that general toxicity affected insulin secretion (Fig. 8 C). The kinetic curves of surface plasmon resonance (SPR) showed that MF286 definitely bound SNAP23, but it barely bound SNAP25, demonstrating that the specific inhibition of SNAP23 resulted in the enhanced insulin secretion (Fig. 8 D).

The addition of MF286, up to 0.01 µM, also increased insulin secretion in the isolated islets in a dose-dependent manner (Fig. 8 E). However, it was less effective at higher concentrations, presumably because of cytotoxicity against the primary β cells. These results indicate that a suitable dose of MF286 increased insulin secretion through SNAP23 inhibition.

To determine the effect of MF286 in vivo, we injected MF286 i.p. into wild-type mice and performed an IPGTT. The blood glucose levels in MF286-injected mice were significantly lower than in PBS-injected mice 30 min after glucose stimulation (Fig. 9 A). Furthermore, the serum insulin levels in the MF286-injected mice were higher than in the uninjected mice (Fig. 9 B). In contrast, ITT revealed similar insulin sensitivities in the mice independent of MF286 treatment (Fig. 9 C). These results suggested that MF286 effectively increased insulin secretion in vivo.

To understand the associated molecular mechanism, we analyzed the effect of MF286 on SNARE complex formation using in vitro GST-pulldown assays. When we added MF286 to a mixture that contained SNAP23, VAMP2, and syntaxin1A, the amount of VAMP2 that bound the syntaxin1A–SNAP23 complex was clearly reduced (Fig. 9 D, left). In contrast, when we added MF286 to a mixture of SNAP25, VAMP2, and syntaxin1A, the amount of VAMP2 that bound the syntaxin1A–SNAP25 complex remained unaltered (Fig. 9 D, right). These results suggest that MF286 specifically inhibits the formation of SNAP23–VAMP2–syntaxin1A complexes and that this inhibition may result in the increased formation of SNAP25–VAMP2–syntaxin1A complexes.

Next, to check whether the effect of MF286 is specific to SNAP23, we performed an IPGTT using BcKO mice. We assumed that if MF286 reduce blood glucose levels through any other different molecules from SNAP23, it would reduce blood glucose even in BcKO mice. However, the blood glucose levels in MF286-injected BcKO mice showed no significant differences compared with PBS-injected BcKO mice (Fig. S4 A). Plasma insulin levels after glucose stimulation also similar between these mice (Fig. S4 B). These results suggest that the increased insulin secretion by MF286 injection is caused by the inhibitory effect of MF286 against only SNAP23.

In peripheral tissues such as the adipose tissue and muscle, glucose is incorporated into cells through the GLUT4 glucose transporter after insulin stimulation. SNAP23 participates in GLUT4 translocation through binding to VAMP2 and syntaxin4 (Kawanishi et al., 2000). To understand the effect of MF286 on the SNAP23, VAMP2, and syntaxin4 complex formation, we performed a GST-pulldown assay. We found no influence of MF286 on the SNARE complex formation among these SNARE proteins (Fig. 9 E). Furthermore, GLUT4 translocation and the uptake of 2-deoxyglucose were not altered in MF286-treated 3T3-L1 adipocytes after insulin stimulation (Fig. 9, F and G). These results suggest that MF286 does not affect GLUT4 translocation in the adipocytes. This is also supported from the previous in vivo ITT, in which glucose uptake was unchanged regardless of MF286 injection.

We also analyzed the effect of MF286 on secretion in the pancreatic acinar cells. The amount of secreted amylase in the MF286-treated acinar cells was significantly decreased (Fig. 9 H), indicating that MF286 inhibits the SNARE complex formation required for ZG fusion in the pancreatic acinar cells.

Finally, we examined whether long-term treatment of MF286 causes adverse effects. MF286 was injected i.p. into wild-type mice once per day for 30 d. Serum biochemical tests showed no differences between PBS- and MF286-injected mice before or after repetitive injection (Fig. S5 A and Table 2). HE staining of various organs also showed no abnormality in MF286-injected mice (Fig. S5 B). These results suggest that MF286 does not cause the adverse effects at least 30 d after injection and may be a safer drug candidate for diabetes.

SNAP23 is essential for exocytosis in various nonneuronal tissues, including the exocrine and endocrine pancreas (Sadoul et al., 1997; Kawanishi et al., 2000; Abonyo et al., 2004; Reales et al., 2005). However, as the function of SNAP23 in regulated secretion in vivo remained unclear, we generated and analyzed both exocrine and endocrine pancreas-specific KO mice.

In the exocrine pancreas, previous studies suggested that in vitro, SNAP23 bound syntaxin2/VAMP2 for primary (ZG-to-plasma membrane) fusion or syntaxin3/VAMP8 for sequential (ZG-to-ZG) fusion (Wang et al., 2004; Weng et al., 2007; Behrendorff et al., 2011). In this study, our results provide direct evidence that SNAP23 promotes both types of fusion in the exocrine pancreas also in vivo. Furthermore, amylase secretion was also decreased in SNAP23 KO parotid exocrine cells, suggesting that SNAP23 is essential for exocytosis in the exocrine system in general.

In the endocrine pancreas-specific KO mice, however, the fusion frequency of predocked insulin granules to the plasma membrane was increased. This result suggests that SNAP23 inhibits the fusion of predocked granules in β cells. The SNARE proteins involved in the secretion of predocked insulin granules are syntaxin1A (or syntaxin4), VAMP2, and SNAP25 (Xie et al., 2015). In particular, the syntaxin1A–VAMP2–SNAP25 complex is specific for the fusion of predocked insulin granules (Ohara-Imaizumi et al., 2007). SNAP23 can also bind syntaxin1A and VAMP2. However, the SNAP23–syntaxin1A–VAMP2 complex is less stable (Montana et al., 2009) and less efficient than the SNAP25–syntaxin1A–VAMP2 complex in mediating proteoliposome fusion or dense-core vesicle fusion (Sørensen et al., 2003; Vites et al., 2008). Accordingly, in β cells, the SNAP23–syntaxin1A–VAMP2 complex seems to decrease frequency of insulin granule-to-plasma membrane fusion by competitive inhibition of SNAP25–syntaxin1A–VAMP2 complex formation, consistent with the fact that the SNAP25–syntaxin1A–VAMP2 complex is not formed before stimulation in β cells (Kasai et al., 2012; Takahashi et al., 2015). Thus, SNAP23 depletion leads to increased usage of SNAP25, which subsequently leads to increased fusion of the granules with plasma membranes (Fig. 10). In contrast, in pancreatic acinar cells, expressing only SNAP23, ZG fusion is mediated by SNAP23–syntaxin2–VAMP2 or SNAP23–syntaxin3–VAMP8 complex (Behrendorff et al., 2011; Thorn and Gaisano, 2012), and, at least, the latter complex can fuse proteoliposome (Xu et al., 2015; Tadokoro et al., 2016). Thus, fusion activity of SNAP23 is lower than SNAP25, but it seems to be sufficient for ZG-to-plasma membrane fusion in pancreatic acinar cells.

SNAP23 and SNAP25 are spatially segregated in neurons (Suh et al., 2010). Our study showed that, in pancreatic β cells, both SNAP23 and SNAP25 localized to the plasma membranes and acted antagonistically on insulin secretion in vivo. Thus, our study revealed differential secretion mechanisms in pancreatic acini and islets as well as functional differences between SNAP23 and SNAP25 in the same β cells. However, the detailed molecular mechanism for fusion inhibition by SNAP23 remains unclear. Thus, further investigation, including in vitro fusion assay, will be required to address this issue.

Because the steady-state serum insulin levels were within the normal range in the β cell–specific KO mice, we reasoned that SNAP23 inhibitor would be less likely to cause hypoglycemia. Thus, they would be safer drug candidates for diabetes mellitus. We screened a library of small compounds and identified several compounds that bound specifically to SNAP23 but not to SNAP25. One of these compounds, MF286, increased insulin secretion in vitro and in vivo (Figs. 8, 9, and 10). MF286 acts on the last stage of insulin secretion, which is distinct from the stages on which other drugs act. Thus, MF286 might be a drug candidate for diabetes that could be used in addition to other drugs.

Additionally, MF286 decreased the amylase secretion from pancreatic acinar cells (Fig. 9 H), suggesting that MF286 might also inhibit the formation of the SNAP23–syntaxin2/3–VAMP2/8 complex. Because SNAP23 is involved in the ectopic fusion of ZGs in alcoholic pancreatitis (Cosen-Binker et al., 2008), we expected that MF286 might also be a drug candidate for pancreatitis.

In conclusion, our results demonstrate that SNAP23 is essential for the fusion of ZGs in the pancreatic acini, but it inhibits the fusion of insulin granules in pancreatic β cells. We also demonstrate that MF286, a SNAP23-binding molecule, increases insulin secretion and improves glucose tolerance in vivo. These results suggest the potential of MF286 as a novel drug candidate for the treatment of diabetes and pancreatitis.

All animal procedures were performed in accordance with the guidelines of the Animal Care and Experimentation Committee of Gunma University and Osaka University. All animals were bred at the Institute of Animal Experience Research of Gunma University and the Institute of Experimental Animal Sciences of Osaka University Medical School.

Snap23 KO mice were generated largely according to previously described methods (Sato et al., 2007). Snap23 genomic clones were isolated from a mouse bacterial artificial chromosome library (RPCI-22; CHORI) using a 0.6-kb mouse Snap23 genomic fragment as a probe. Within the targeting vector, a single loxP site was inserted into intron 2, and the splice acceptor–internal ribosomal entry site–β-geo-poly(A) cassette with two flippase recombinase target sites and one loxP site was inserted into intron 5 (Fig. 1 A). This construct was electroporated into embryonic stem cells. Homologous recombinant embryonic stem cells were identified by Southern blot and injected into C57BL/6 to obtain chimeric mice for generation of Snap23^geo/+ mice. To obtain Snap23^−/− mice and Snap23 floxed mice (Snap23^{floxed/floxed}), we crossed Snap23^geo/floxed mice with CMV-Cre and Act-Flp-e mice (The Jackson Laboratory), respectively. We crossed Snap23 floxed mice with Elastase-Cre (Hashimoto et al., 2008), RIP-Cre (Herrera, 2000; Kitamura et al., 2009), or Pdx1-Cre (Gu et al., 2002) mice to generate pancreatic acinar cell– or β cell–specific Snap23 KO mice. Each mouse line was backcrossed at least 10 times onto a C57BL/6 background. The genotypes of the mice were identified by PCR using the following primers: primer 1 (5′-CTGGGGAATGTGCGTTTGGATGATG-3′), primer 2 (5′-CCCCTTTCATCATGCTTCAAATGCAACC-3′), primer 3 (5′-TGTTCCTGGATTGAGCTCAGGTGGT-3′), primer 4 (5′-AGGTTCGTTCTCTCATGGA-3′), and primer 5 (5′-TCGACCAGTTTAGTTACCC-3′).

Primers 1 and 2, 2 and 3, and 4 and 5 were used for the floxed allele, the null allele, and the cre gene, respectively (Fig. 1 A).

Blood samples of each genotype mouse were collected from tail vain after fasting for 18 h. Serum was isolated by centrifugation (3,000 rpm for 15 min). Serum total protein, albumin, blood urea nitrogen, creatinine, sodium, potassium, chloride, calcium, inorganic phosphorus, aspartate aminotransferase, alanine aminotransferase, lactate dehydrogenase, amylase, total cholesterol, triglycerides, high-density lipoprotein (HDL) cholesterol, total bilirubin, and glucose levels were quantified using a 7180 Automatic Analyzer (Hitachi High-Technologies) by the Nagahama Life Science Laboratory.

For HE staining, 8–12-wk-old mice were intracardially perfused with 3% PFA in 0.1 M phosphate buffer, pH 7.4. The pancreas was removed from each mouse and subjected to additional fixation in the same fixative for 2 h. After fixation, the pancreas was dehydrated in ethanol and embedded in paraffin. Paraffin-embedded pancreatic sections (4 µm) were stained with a solution of HE. For immunohistochemistry, paraffin-embedded sections of human pancreas were purchased from BioChain Institute, Inc. The sections were labeled using primary and secondary antibodies as described in the Antibodies section after antigen retrieval (autoclaved in citrate buffer). For immunofluorescence microscopy, cryoprotection was performed after fixation by incubating the tissues consecutively in 4, 10, 15, and 20% sucrose in 0.1 M phosphate buffer, pH 7.4. The tissues were embedded in O.C.T. compound (Sakura) and then frozen in liquid nitrogen–chilled isopentane. The frozen sections were cut using a CM3000 cryostat (Leica Biosystems) and labeled with primary and secondary antibodies as described in the Antibodies section. Confocal images were obtained using an FV1000D laser-scanning microscope (Olympus) with a UPlanSApo 60× objective lens (NA 1.35; Olympus).

The anti-SNAP23 polyclonal antibody was raised against a purified recombinant His6-SNAP23 protein according to previously described protocols (Chen et al., 1999). Rabbits were immunized with purified His6-SNAP23. The resulting serum was purified by passage through an affinity column with GST-SNAP25. The serum was further purified by affinity chromatography with GST-SNAP23.

The following primary antibodies were used: SNAP25 (mouse; SMI-81; Covance), SNAP25 (rabbit; 5308; Cell Signaling Technology), syntaxin1A (mouse; S0664; Sigma-Aldrich), VAMP2 (mouse; 104211; Synaptic Systems), VAMP2 (rabbit; 13508; Cell Signaling Technology), GP2 (mouse; 2F11-C3; MBL International Corporation), GAPDH (mouse; CB1001; EMD Millipore), and GLUT4 (provided by H. Shibata, Institute for Molecular and Cellular Regulation, Gunma University, Gunma, Japan; Shibata et al., 1995). The anti-GST rabbit polyclonal antibody was raised against a purified GST protein, and the antiserum was affinity purified with a GST column. For secondary antibodies, EnVision+ System HRP-labeled polymer anti-rabbit and anti-mouse (Dako) and Alexa Fluor 488–, 568–, and 594–labeled donkey anti–rabbit and anti–mouse IgG (Molecular Probes) were used.

Tissues or cells were homogenized in lysis buffer (80 mM Tris-HCl, pH 6.8, and 2% SDS) containing a protease inhibitor cocktail (Nacalai Tesque, Inc.); the lysates were subsequently boiled and centrifuged at 20,000 g for 10 min. The supernatants were used for SDS-PAGE. The antibodies noted in the previous section were used as primary antibodies. HRP-labeled donkey anti-mouse and anti-rabbit antibodies (Jackson ImmunoResearch Laboratories, Inc.) were used as secondary antibodies. An Immobilon Western Chemiluminescent HRP Substrate kit (EMD Millipore) was used for detection. Chemiluminescent images were obtained using an automatic film developer.

For electron microscopy, 8–12-wk-old control, AcKO, or PcKO mice were used. CCK (0.25 µg/kg weight) or glucose (1 g/kg weight) was administered by i.p. injection. 30 min after the injections, the mice were perfused intracardially with 2.5% glutaraldehyde and 2% PFA in 0.1 M cacodylate buffer, pH 7.4. The pancreas was removed from each mouse and fixed in the same fixative and 0.1% OsO₄, after which it was further fixed in 0.5% uranyl acetate in H₂O, dehydrated, and embedded in Quetol 812 (Nisshin EM). Ultrathin sections were cut using an ultra-microtome (Reichert Jung). Electron micrographs were taken using a JEM-1010 (JEOL).

Acinar cells were isolated from control and AcKO mice as previously described (Nemoto et al., 2001). The isolated acinar cells were incubated with or without 100 pM CCK for 1 h at 37°C. Secreted amylase from the acinar cells was measured using a QuantiChrom Alpha-Amylase Assay kit (BioAssay Systems) according to the manufacturer’s instructions.

For extraction of lipids, stools were homogenized with chloroform/methanol (2:1). The homogenate was centrifuged, and the lipid-containing chloroform layer was collected. The solvent was mixed with 0.36 M CaCl₂ in methanol, and the chloroform layer was collected again after centrifugation. Chloroform was evaporated, and remaining lipids were used for quantification. Triglycerides were measured using a Triglyceride E-Test Wako (Wako Pure Chemical Industries) according to the manufacturer’s instructions.

Primary parotid acinar cell culture was performed as previously described (Fujita-Yoshigaki et al., 2005). Parotid acinar cells were isolated from Snap23 floxed mice (Snap23^{floxed/floxed}) and infected with LacZ- or Cre-expressing adenovirus for 48 h. After infection, cells were stimulated with 1 µM IPR and 10 µM MeCh for 15 min at 37°C. Secreted amylase in the supernatant was quantified using colorimetric assay. Total RNA was extracted from acinar cells using a NucleoSpin RNA kit (Macherey-Nagel) according to the manufacturer’s instructions. First-strand cDNA was synthesized by random primers (6-mers) and reverse transcription (Takara Bio Inc.). Real-time PCR was performed using ViiA7 (Applied Biosystems). Universal Probes 13 and 78 (Roche) were used for SNAP23 and glucose-6-phosphate dehydrogenase X-linked (G6PDX), respectively. The primers used for PCR were as follows: SNAP23 sense (5′-CACTATGCTGGATGAGCAAGG-3′), SNAP23 antisense (5′-CAACACTTGTTGAGTTCTGTTAAAGTC-3′), G6PDX sense (5′-GAAAGCAGAGTGAGCCCTTC-3′), and G6PDX antisense (5′-CATAGGAATTACGGGCAAGA-3′).

The isolation of pancreatic acinar cells and β cells was described previously (Nemoto et al., 2001; Takahashi et al., 2002). ZG exocytosis in the control and AcKO acinar cells or insulin granule exocytosis in the control and PcKO β cells was visualized using a solution that contained 0.5 or 0.7 mM SRB as a fluid-phase tracer. Two-photon excitation imaging for pancreatic acinar cells was performed using a laser-scanning microscope (A1R MP+; Nikon) with an XLPLan 25× objective lens (NA 1.05; Olympus) and a femtosecond laser (Mai Tai eHP DeepSee; Spectra-Physics). For pancreatic β cells, two-photon imaging was performed using an inverted laser-scanning microscope (FV1000 and IX81; Olympus) equipped with a water-immersion objective lens (UPlanApo 60× W/IR; NA 1.2; Olympus) and a femtosecond laser (Mai Tai; Spectra-Physics). Exocytosis was measured in response to 100 pM CCK or 20 mM glucose within arbitrary areas of the isolated acinar cells and islets. Increases in the cytosolic Ca²⁺ concentration were measured using the Fura-2 AM (K_Ca: 0.2 µM) or Fura-2 FF (K_Ca: 40 µM) Ca²⁺ indicator and were reported as (F₀ – F)/F₀, where F₀ and F represent the resting and post-stimulation fluorescence levels, respectively.

The IPGTT, ITT, and measurement of plasma insulin concentration were performed as described previously (Gomi et al., 2005). Blood glucose concentrations in control, BcKO, or PcKO mice were measured at 15, 30, 60, and 120 min after glucose or insulin (Humalin R; Eli Lilly and Company) injection (1 g/kg or 0.75 U/kg weight) with ACCU-CHEK (Roche) or Glutest Neo Super (Sanwa Kagaku Kenkyusho). To measure plasma insulin concentration, plasma samples were obtained from control or BcKO mice at 0, 15, 30, and 60 min after glucose injections. An LBIS mouse insulin ELISA kit (Shibayagi) was used according to the manufacturer’s instructions.

Pancreatic islets were isolated from control and BcKO mice by collagenase digestion, as previously described (Ohara-Imaizumi et al., 2004). For the batch incubation assay, 10 size-matched islets were preincubated for 30 min in Krebs-Ringer buffer (KRB) containing 110 mM NaCl, 4.4 mM KCl, 1.45 mM KH₂PO₄, 1.2 mM MgSO₄, 2.3 mM calcium gluconate, 4.8 mM NaHCO₃, 2.2 mM glucose, 10 mM Hepes, pH 7.4, and 0.3% BSA and then transferred into 1 ml KRB containing 2.2 or 16.7 mM glucose. After 30-min incubation, 200 µl supernatant was recovered as the secreted insulin sample. For the perfusion assay, 40 size-matched islets were housed in a small chamber and perifused with KRB containing 2.8 mM glucose for 30 min at a flow rate of 0.5 ml/min, and insulin release was stimulated by 22 mM glucose for 30 min. After the experiments, islets were solubilized by 1% Triton X-100 and sonicated on ice to recover the total cellular content of insulin. The secreted insulin and total cellular content of insulin were measured using an insulin ELISA kit (Morinaga Institute of Biological Science, Inc.).

The isolated islets from control and BcKO mice were dispersed and cultured on fibronectin-coated high-refractive index cover glass (Olympus) in RPMI 1640 medium (Gibco) supplemented with 10% FBS (Gibco), 200 U/ml penicillin, and 200 µg/ml streptomycin at 37°C in an atmosphere of 5% CO₂ (Ohara-Imaizumi et al., 2007). To label insulin secretory granules, β cells were infected with recombinant adenovirus Adex1CA insulin-GFP. Experiments were performed 18–24 h after the final infection. A total internal reflection system with a high-aperture objective lens (Apo 100× OHR; NA 1.65; Olympus) was used as previously described (Ohara-Imaizumi et al., 2007). In brief, β cells expressing insulin-GFP on the cover glass were mounted in an open chamber and incubated for 30 min with KRB. Cells were then transferred to a thermostat-controlled stage (37°C) and stimulated with 22 mM glucose. TIRF images in live cells were acquired every 300 ms after glucose stimulation. The data analysis was performed using Metamorph software (Universal Imaging). The number of fusion events per cell was manually counted. The exocytotic fusion events in each corresponding cell were counted during a time course.

The in vitro SNARE binding assay was performed as previously described (Pevsner et al., 1994). Full-length human Syntaxin1A cDNA, inserted into the pET-41 Ek/LIC vector (provided by J. Mima, Institute for Protein Research, Osaka University, Osaka, Japan), was used to generate the recombinant GST-Syntaxin1A protein. The cytoplasmic region of mouse syntaxin4 inserted into the pGEX-6P-1 vector was used to generate the recombinant GST–syntaxin4 protein. Full-length mouse SNAP25 cDNA and cDNA containing the cytoplasmic region of mouse VAMP2 were subcloned into the pQE32 vector (QIAGEN) to generate the recombinant His-SNAP25 and His-VAMP2 proteins, respectively. All recombinant proteins, except GST-Syntaxin4, were expressed in Rosetta2 Escherichia coli as His6-tagged fusion proteins and purified using Ni-NTA agarose (QIAGEN) according to the manufacturer’s instructions. GST-Syntaxin4 was expressed in Rosetta2 E. coli as a GST-tagged fusion protein and purified with Glutathione Sepharose 4B (GE Healthcare) according to the manufacturer’s instructions.

For the binding competition experiment, His-SNAP23 (0, 0.6, 5, and 10 µM), His-Stx1A, and His-VAMP2 (0.6 µM each) were mixed in buffer (150 mM NaCl, 20 mM Tris-HCl, pH 7.5, 1 mM EDTA, and 5% glycerol) and incubated for 2 h at 4°C. The mixture was combined with GST-SNAP25 (0.6 µM) and Glutathione Sepharose 4B (GE Healthcare) and rotated overnight at 4°C. The glutathione beads were washed three times in wash buffer (300 mM NaCl, 20 mM Tris-HCl, pH 7.5, 1 mM EDTA, 5% glycerol, and 0.1% Triton X-100) and subjected to microcentrifugation for 30 s at 4°C. The beads were boiled and analyzed by SDS-PAGE and immunoblotting.

For the MF286 experiment, His-SNAP23 or His-SNAP25 was incubated with ethanol or MF286 (2 µM) for 1 h at 4°C. Next, 0.6 µM each of His-SNAP23 or His-SNAP25, His-VAMP2, and GST-Syntaxin1A or GST-Syntaxin4 was combined in buffer (150 mM NaCl, 20 mM Tris-HCl, pH 7.5, 1 mM EDTA, and 5% glycerol) with ethanol or 2 µM MF286 and rotated with Glutathione Sepharose 4B (GE Healthcare) overnight at 4°C. The glutathione beads were washed three times in wash buffer and subjected to microcentrifugation for 30 s at 4°C. The beads were boiled and analyzed by SDS-PAGE and immunoblotting.

Immunoprecipitation was performed as previously described (Oh and Thurmond, 2009). Isolated islets from control and BcKO mice or MIN6 cells (provided by J. Miyazaki, Osaka University, Osaka, Japan) were preincubated for 30 min in glucose-free modified Krebs ringer bicarbonate buffer (5 mM KCl, 120 mM NaCl, 15 mM Hepes, pH 7.4, 24 mM NaHCO₃, 1 mM MgCl₂, 2 mM CaCl₂, and 1 mg/ml BSA) with 1 mM N-ethylmaleimide. The islets were additionally incubated with 25 mM glucose for 15 min. The islets or MIN6 cells were lysed in NP-40 lysis buffer (1% NP-40, 25 mM Hepes, pH 7.4, 10% glycerol, 137 mM NaCl, and protease inhibitor cocktail), and the lysates were subsequently cleared by microcentrifugation for 10 min at 4°C. A total of 10 mg of cleared lysate was used for immunoprecipitation with the appropriate antibodies. The precipitated products were boiled and analyzed by SDS-PAGE and immunoblotting. Rabbit TrueBlot HRP-anti rabbit antibody (Rockland Immunochemicals) was used for the detection of SNAP25 or SNAP23 in the immunoprecipitate from the lysate.

MIN6 cells were cultured in DMEM (Wako Pure Chemical Industries) supplemented with 10% FCS and 0.0005% β-mercaptoethanol at 37°C under a 5% CO₂/95% air atmosphere.

For SNAP23 or SNAP25 knockdown, the following Ambion silencer siRNAs (Thermo Fisher Scientific) were transfected with Lipofectamine RNAiMAX (Invitrogen): control siRNA (AM4611), SNAP23 siRNA (64778; 5′-GGCAUGGACCAAAUAAAUATT-3′), and SNAP25 siRNA (151786; 5′-GCAACAACUACGCAUGCUCTT-3′).

The hGH-expressing MIN6 cells were washed three times with KRB. After washing, the cells were incubated with KRB plus 0.3% BSA and stimulated with KRB plus 30 mM KCl. At the experimental endpoint, 1 ml of chilled 1% (vol/vol) NP-40 was added, and the samples were sonicated (UR-20P; Tomy Seiko Co.) on ice. Secreted hGH and the total cellular content of hGH were measured using an hGH ELISA kit (Roche).

Chemical array screening was performed as previously described (Hagiwara et al., 2010; Zimmermann et al., 2013). Eight array slides that included 23,275 compounds were incubated with 2 µM GST-SNAP23 or 500 µl GST-SNAP25 solution for 1 h at 30°C in the following buffer: 10 mM Tris-HCl, 150 mM NaCl, 0.05% Tween 20, and 1% skim milk, pH 8. After washing, the array slides were incubated with anti-GST antibody for 1 h at 30°C. This incubation was followed by another wash step and incubation with a Cy5-conjugated secondary antibody for 1 h at 30°C. After a final wash step, the slides were scanned at 635 nm using a GenePix 4300A microarray scanner (Molecular Devices). The compounds that bound SNAP23 were identified by the fluorescence intensity of each spot.

MIN6 cells that had been cultured in 96-well plates were incubated overnight with chemical compounds (3, 10, and 30 µg/ml). MIN6 cell viability was measured after the compound treatments using Cell Count Reagent SF (Nacalai Tesque, Inc.) according to the manufacturer’s instructions. Isolated islets from wild-type mice were incubated with MF286 (0, 0.001, 0.01, 0.1, 1, and 10 µM) for 1 h. To measure secreted insulin, the islets or MIN6 cells were washed three times with KRB plus 2.2 or 2.8 mM glucose. After washing, they were incubated with KRB and subsequently stimulated with KRB plus 22 or 13.9 mM glucose for 1 h. A rat Insulin RIA kit (EMD Millipore) or an LBIS mouse insulin ELISA kit (Shibayagi) was used to measure insulin according to the manufacturer’s instructions.

Isolated acinar cells from wild-type mice were incubated with MF286 (37.5 µM) or 300 nM tetanus toxin (Sigma-Aldrich) for 1 h at 37°C, as previously described (Rosado et al., 2005). Cells were then stimulated by 100 pM CCK for 1 h at 37°C. Secreted amylase from the acinar cells was measured using a QuantiChrom Alpha-Amylase Assay kit (BioAssay Systems) according to the manufacturer’s instructions.

For studying the in vivo effect, MF286 was injected i.p. (0.4 or 4 mg/kg weight) into 8-wk-old female C57BL/6 mice. IPGTT, ITT, and plasma insulin were measured at 15 or 30 min after MF286 injection as described in IPGTT, ITT, and measurement of plasma insulin concentration.

MF286 and DS565-8 (10 mM each) were spotted and immobilized onto the sensor chip (gold-coated glass slide; PBA-CS-03-5; Plexera) by UV cross-linking. Purified His-SNAP23 or His-SNAP25 protein in running buffer (150 mM NaCl, 20 mM Tris-HCl, pH 7.5, 1 mM EDTA, and 5% glycerol) was loaded at 30°C. The changes in reflected light intensity were recorded and displayed as a sensorgram using a PlexArray HT Analyzer (Plexera).

The adipogenesis of 3T3-L1 fibroblasts was induced by treatment with insulin, dexamethasone, and isobutylmethylxanthine as described previously (Kawanishi et al., 2000). 3T3-L1 adipocytes were incubated with or without MF286 (37.5 µM) for 2 h. The 2-deoxyglucose incorporated into 3T3-L1 adipocytes was measured using a 2-DG Uptake Measurement kit (Cosmo Bio) according to the manufacturer’s instructions.

Fig. S1 shows coimmunostaining of SNAP23 and insulin, glucagon, and somatostatin in pancreatic islets from control, BcKO, and PcKO mice. Fig. S2 shows coimmunostaining of SNAP25 and insulin, glucagon, and somatostatin in pancreatic islets from control, BcKO, and PcKO mice. Fig. S3 shows glucose tolerance and insulin secretion in PcKO mice. Fig. S4 shows IPGTT and plasma insulin concentration in BcKO mice with or without MF286. Fig. S5 shows the effect of long-term administration of MF286. Video 1 shows two-photon imaging of exocytotic events in control and AcKO acinar cells. Video 2 shows two-photon imaging of exocytotic events in control and PcKO islets.

We thank M. Takano, T. Akuzawa, K. Inami, T. Horie, Y. Okada, R. Hirai, H. Kobayashi, H. Hata, N. Atik, M. Taniguchi, A. Goto, H. Togawa, A. Watanabe, Y. Noguchi, A Koizumi, T. Saito, and K. Honda for technical assistance; D. C. Thurmond, M. Sato, T. Sobajima, M. Inoue, and T. Hara for advice; J. Miyazaki for providing the MIN6 cells; J. Mima for advice and for providing the human syntaxin1A cDNA construct; and H. Shibata for providing the GLUT4 antibody. The PlexArray HT Analyzer was lent by the Division of Innovative Research for Drug Development, Osaka University Institute for Academic Initiatives.

This work was supported by Grants-in-Aid for the 21st Century Center of Excellence Program of the Japanese Ministry of Education, Culture, Sports, Science, and Technology (to K. Sato and A. Harada), Grants-in-Aid for Scientific Research from the Japan Society for the Promotion of Science (KAKENHI grants JP24390046 to A. Harada and JP26870335 to M. Kunii),the Daiwa Securities Health Foundation, the Suzuken Memorial Foundation (to T. Sato), the Daiichi Sankyo Foundation of Life Science, Novartis Foundation (Japan) for the Promotion of Science (to A. Harada), The Uehara Memorial Foundation (to S. Yoshimura), and the joint research program of the Institute for Molecular and Cellular Regulation, Gunma University.

The authors declare no competing financial interests.

Author contributions: M. Kunii, M. Ohara-Imaizumi, N. Takahashi, M. Kobayashi, R. Kawakami, Y. Kondoh, T. Shimizu, S. Simizu, B. Lin, K. Nunomura, K. Aoyagi, M. Ohno, M. Ohmuraya, T. Sato, S. Yoshimura, K. Sato, R. Harada, Y.-J. Kim, H. Osada, T. Nemoto, H. Kasai, T. Kitamura, S. Nagamatsu, and A. Harada conducted the experiments. A. Harada supervised the project. M. Kunii, R. Harada, and A. Harada wrote the manuscript.