Cell death-inducing DNA fragmentation factor A-like effector A and fat-specific protein 27β coordinately control lipid droplet size in brown adipocytes

Adipose tissue stores neutral lipids and is a major metabolic organ involved in regulating whole-body energy homeostasis. Triacylglycerol is stored as unilocular large lipid droplets (LDs) in white adipocytes and as multilocular small LDs in brown adipocytes. Proteins of the cell death-inducing DNA fragmentation factor A-like effector (Cide) family include CideA, CideB, and fat-specific protein of 27 (FSP27). Of these, FSP27 has been shown to play a crucial role in the formation of unilocular large LDs in white adipocytes. However, the mechanisms by which brown adipocytes store small and multilocular LDs remain unclear. An FSP27 isoform, FSP27β, was recently identified. We herein report that CideA and FSP27β are mainly expressed in brown adipose tissue and that FSP27β overexpression inhibits CideA-induced LD enlargements in a dose-dependent manner in COS cells. Furthermore, RNAi-mediated FSP27β depletion resulted in enlarged LDs in HB2 adipocytes, which possess the characteristics of brown adipocytes. Brown adipocytes in FSP27-knock-out mice that express CideA, but not FSP27β, had larger and fewer LDs. Moreover, we confirmed that FSP27β and CideA form a complex in brown adipose tissue. Our results suggest that FSP27β negatively regulates CideA-promoted enlargement of LD size in brown adipocytes. FSP27β appears to be responsible for the formation of small and multilocular LDs in brown adipose tissue, a morphology facilitating free fatty acid transport to mitochondria adjacent to LDs for oxidation in brown adipocytes.


Adipose tissue stores neutral lipids and is a major metabolic organ involved in regulating whole-body energy homeostasis.
Triacylglycerol is stored as unilocular large lipid droplets (LDs) in white adipocytes and as multilocular small LDs in brown adipocytes. Proteins of the cell death-inducing DNA fragmentation factor A-like effector (Cide) family include CideA, CideB, and fat-specific protein of 27 (FSP27). Of these, FSP27 has been shown to play a crucial role in the formation of unilocular large LDs in white adipocytes. However, the mechanisms by which brown adipocytes store small and multilocular LDs remain unclear. An FSP27 isoform, FSP27␤, was recently identified. We herein report that CideA and FSP27␤ are mainly expressed in brown adipose tissue and that FSP27␤ overexpression inhibits CideA-induced LD enlargements in a dose-dependent manner in COS cells. Furthermore, RNAi-mediated FSP27␤ depletion resulted in enlarged LDs in HB2 adipocytes, which possess the characteristics of brown adipocytes. Brown adipocytes in FSP27-knock-out mice that express CideA, but not FSP27␤, had larger and fewer LDs. Moreover, we confirmed that FSP27␤ and CideA form a complex in brown adipose tissue. Our results suggest that FSP27␤ negatively regulates CideA-promoted enlargement of LD size in brown adipocytes. FSP27␤ appears to be responsible for the formation of small and multilocular LDs in brown adipose tissue, a morphology facilitating free fatty acid transport to mitochondria adjacent to LDs for oxidation in brown adipocytes.
Adipose tissue is specialized for storing neutral lipids and is one of the main metabolic organs involved in the regulation of whole-body energy homeostasis. There are two types of adipocytes (1): white adipocytes, which are energy-storing adipocytes, and brown adipocytes, which are energy-dissipating adipocytes. The former stores triacylglycerol (TAG) 2 effectively as unilocular large lipid droplets (LD) and supplies free fatty acids (FFAs) and glycerol to other tissues during fasting by hydrolyzing stored TAG. The latter consumes TAG stored as multilocular LD inordertogenerateheatthroughthemitochondrialoxidativephosphorylation of FFA and subsequent uncoupling in response to cold. The different morphologies of LD between these two types of adipose cells may reflect their unique characteristics.
Recent studies clarified that cell death-inducing DNA fragmentation factor A-like effector (Cide)-family proteins play a crucial role in lipid and energy metabolism including lipolysis, lipid oxidation, and LD formation (2). CideA is strongly expressed in brown adipocytes in mice (3), whereas CideB is mainly expressed in the liver and kidney (4). Fat-specific protein of 27 (FSP27) (CideC in humans) is abundantly expressed in adipose tissue and contributes to the formation of unilocular LD in white adipocytes (5)(6)(7). Although CideA and FSP27 are both involved in the formation of large LD (5)(6)(7)(8)(9)(10)(11)(12), the molecular mechanisms by which brown adipocytes form multilocular small LD despite the abundant expression of CideA currently remain unknown. A new isoform of FSP27, FSP27␤, was recently identified and is a molecule to which 10 amino acids were added to the amino-terminal domain of the conventional type of FSP27, designated as FSP27␣ (13). The expression of FSP27␤ in the liver is regulated by the liver-enriched transcription factor cyclic-AMPresponsive-element-binding protein H (CREBH) (13).
In the present study we investigated the mechanisms by which Cide-family proteins regulate the size and structure of LD in adipocytes and revealed that CideA and FSP27␤ regulate the formation of small and multilocular LD in brown adipocytes in a coordinated manner.

Expression of FSP27␣, FSP27␤, and CideA in white and brown adipose tissues
We initially investigated the mRNA expression of FSP27␣, FSP27␤, and CideA in the epididymal white adipose tissue (eWAT) and brown adipose tissue (BAT) of mice using a quantitative RT-PCR method with primers that specifically recognize each isoform. FSP27␣ was mainly expressed in eWAT. In contrast, FSP27␤ and CideA were strongly expressed in BAT (Fig. 1A). We then confirmed the protein expression of these family members using an immunoblot analysis with an anti-CideA antibody and anti-FSP27 antibody that recognized the FSP27␣ and FSP27␤ isoforms. FSP27␣ was mainly expressed in eWAT, whereas FSP27␤, which is 10 amino acids longer than FSP27␣, was dominantly expressed in BAT (Fig. 1B). The CideA protein was specifically expressed in BAT (Fig. 1B). These results were consistent with those obtained on mRNA expression and suggest that FSP27␣ is expressed in eWAT, whereas FSP27␤ and CideA are expressed in BAT.

Effects of the overexpression of FSP27␣, FSP27␤, and CideA on the formation of LD in COS cells
We examined the effects of FSP27␣, FSP27␤, and CideA on the formation of LD in COS cells by overexpressing these proteins using a pIRES2-DSRed2 vector. We confirmed each protein expression in COS cells by immunoblot analysis. The expression level of FSP27␣ seems to be more abundant than FSP27␤ ( Fig. 2A). The different expression levels between FSP27␣ and FSP27␤ may reflect the altered affinity of anti-FSP27 antibody to FSP27␣ and FSP27␤ due to the conformational difference between these two isoforms. In microscopic analysis, COS cells expressing these proteins were identified by monitoring the simultaneously expressed fluorescence marker DSRed. The overexpression of FSP27␣ resulted in the formation of large LD, as was reported previously (Fig. 2, B and C). At the same time, intracellular LD number was decreased significantly (Fig. 2, B and D). The overexpression of CideA also led to the formation of large LD and the decrease of intracellular LD number. However, LD sizes were smaller, and LD numbers were increased in cells overexpressing FSP27␤ than in control cells (Fig. 2, B-D). In the overexpression of FSP27␤, the fluorescence marker protein DSRed2 was localized on the surface of LD (Fig. 2B); however, the reason for this localization of DSRed2 currently remains unknown. We then overexpressed FSP27␣, FSP27␤, and CideA in COS cells using a pcDNA3.1 vector and examined the intracellular localization of these molecules by immunofluorescence microscopy using an anti-FSP27 antibody and anti-CideA antibody. FSP27␣ and CideA both localized on the surface of large LD (Fig. 2E). FSP27␤ also localized around small LD (Fig. 2E). These results suggest that these three proteins are involved in LD formation on the LD surface.
A previous study reported that overexpression of COOHterminal GFP-tagged FSP27␤ increased total cellular TAG content and LD size in COS cells (13). Therefore, to investigate the effects of FSP27␤ as a fusion protein in the LD formation, we overexpressed FSP27␣, FSP27␤, and CideA in COS cells as a fusion protein with the fluorescence protein DSRed-Monomer using the pDSRed-Monomer-C1 vector and examined the LD sizes and numbers. Immunofluorescence microscopy revealed that overexpression of FSP27␣ and CideA as a fusion protein with DSRed actually resulted in the large LD formation, although the degree of enlargement was slightly smaller than in the single expression (Fig. 3, A and B). Overexpression of

Proposed mechanism for LD formation in brown adipocytes
FSP27␤ as a fusion protein with DSRed also increased the LD size very slightly compared with control even though the LD size was much smaller than FSP27␣ and CideA (Fig. 3, A and B). This is inconsistent with the result that overexpression of FSP27␤ solely decreased the LD size (Fig. 2, B and C). We assume this is because DSRed protein attached at the aminoterminal region could modify the original function of FSP27␤. LD numbers were also decreased in cells overexpressing FSP27␣ and CideA as a fusion protein with DSRed (Fig. 3C). Three proteins were actually localized on the LD surface (Fig.   3A). These data clearly indicate that FSP27␣ and CideA, but not FSP27␤, promote large LD formation and decreased intracellular LD number.

FSP27␤ inhibits the CideA-induced enlargement of LD in COS cells
Our results suggest that the main isoforms of the Cide family expressed in BAT are FSP27␤ and CideA. Thus, to reconstitute the condition of BAT, we overexpressed FSP27␤ and CideA in COS cells using the pIRES2-DSRed2 and pcDNA3.1 vectors, respectively. We recognized cells overexpressing FSP27␤ by the fluorescence marker DSRed and those overexpressing CideA by immunofluorescence using an anti-CideA antibody. As shown in Fig. 4, A and B, the overexpression of CideA solely resulted in the formation of large LD. The overexpression of FSP27␤ did not induce large LD (Fig. 4, A and B). Furthermore, the simultaneous overexpression of FSP27␤ and CideA inhibited the CideA-induced large LD formation (Fig. 4, A and B). Co-overexpression of FSP27␤ and CideA increased intracellular LD numbers compared with CideA alone (Fig. 4, A and B). These results suggest that FSP27␤ dominantly inhibits the CideA-induced enlargement of LD and increases LD number. In addition, this inhibition of LD enlargement by the expression of FSP27␤ using the pIRES2-DSRed2 vector was dose-dependent (Fig. 4C), although the control pIRES2-DSRed2 vectors showed no effect on LD size (Fig. 4D). In the previous study, we produced the mutant of FSP27␣ in which the negatively charged acidic polar amino acids (Asp-215, Glu-218, Glu-219, and Glu-220), which are supposed to be important for dimer formation, were replaced by noncharged polar amino acids (Asn-215, Gln-218, Gln-219, and Gln-220) and revealed that these negatively charged acidic polar amino acids (Asp-215, Glu-218, Glu-219, and Glu-220) were indispensable for the function of FSP27␣ to enlarge LD (12). Therefore, we introduced the same mutations to FSP27␤ to decrease the function of FSP27␤ and examined the inhibitory effect of FSP27␤ to CideA-induced LD enlargement. We found that the effect of the mutant FSP27␤ to inhibit the enlargement of LD size promoted by CideA was diminished compared with wild-type FSP27␤ (Fig. 4E), suggesting that FSP27␤ is actually involved in the regulation of LD size. Next, to investigate the intracellular localization of FSP27␤ and CideA in cells expressing both proteins in more detail, we overexpressed FSP27␤ and CideA in COS cells using the pDSRed-Monomer-C1 and pcDNA3.1 vectors, respectively. Immunofluorescence microscopy revealed that the red fluorescence of DSRed fused with FSP27␤ showed the same distribution to the fluorescence of CideA recognized by immunofluorescence using the anti-CideA antibody (Fig. 4F). These results suggest that FSP27␤ inhibits the function of CideA that enlarges LD at the same region on the LD surface.

Depletion of FSP27␤ resulted in the enlargement of LD in brown adipocytes
To confirm that FSP27␤ plays an inhibitory role in CideAinduced LD enlargements in brown adipocytes, we depleted the expression of FSP27␤ in the brown adipocyte cell line, HB2, which was established from the preadipocytes of mouse BAT (14). After the induction of adipogenesis, HB2 cells showed the morphological characteristics of brown adipocytes, including the formation of multilocular small LD in the cytoplasm. These cultured brown adipocytes also expressed FSP27␤, CideA, and FSP27␣ (Fig. 5A). In these cells almost similar amounts of proteins in FSP27␣ and FSP27␤ were expressed. This is because HB2 cells are the established cultured adipocytes, and their characters as brown fat cells are thought to get weaker compared with their original brown fat cells in which FSP27␤ is overwhelmingly expressed. The results of the immunoblot analysis revealed that the knockdown of FSP27 using siRNA markedly decreased endogenous FSP27␣ and -␤ without affecting the expression of CideA (Fig. 5A). The decreased expression of FSP27␣ and -␤ enlarged LD in HB2 brown adipocytes (Fig. 5, B and C). In addition, many cells presented unilocular LD in knockdown of FSP27, although almost all control cells displayed multilocular LD pattern (Fig. 5D). Given that the function of FSP27␣ is the formation of large and unilocular LD, the result obtained by the knockdown of FSP27 was due to the decrease of FSP27␤. We also examined the size of LD in the BAT of FSP27 knock-out (KO) mice. FSP27 was not expressed in eWAT and BAT in FSP27 homo KO mice, although CideA expression levels in BAT did not change among wild-type, FSP27 hetero KO, and FSP27 homo KO mice (Fig.  6A). As demonstrated previously, the size of LD markedly decreased in the eWAT of FSP27 homo KO mice (Fig. 6, B and  C). In contrast, the size of LD was markedly larger in the BAT of FSP27 homo KO mice than in that of wild-type or hetero KO mice (Fig. 6, B and C). The number of LD in the BAT of FSP27 homo KO mice decreased, and the formation of LD shifted more toward unilocular LD (Fig. 6, B and D). These results suggest that FSP27␤ plays an inhibitory role in the CideA-induced enlargement of LD in brown adipocytes.

FSP27␤ inhibits the homo dimerization of CideA in COS cells and actually forms a complex with CideA in brown adipocytes
Previous studies proposed that not only FSP27␣, but also CideA, promote LD growth by forming a homodimer on the contact site of two contiguous LD and inducing their fusion (15)(16)(17). Because FSP27␤ inhibited the CideA-induced enlargement of LD, FSP27␤ may bind to CideA, resulting in the inhibition of the homodimer of CideA and subsequent growth of LD in brown adipocytes. Thus, we investigated whether FSP27␤ inhibited the homodimer formation of CideA in COS cells. We overexpressed CideA tagged with human c-MYC (CideA-MYC) or tagged with human influenza hemagglutinin (HA) (CideA-HA) using the pcDNA3.1 vector. CideA-MYC was co-immunoprecipitated with CideA-HA using the antibody to HA from the detergent extracts of COS cells expressing both CideA-MYC and CideA-HA, suggesting that CideA actually forms a homodimer (Fig. 7A). Additional overexpression of FSP27␤ in cells expressing both CideA-MYC and CideA-HA resulted in the co-immunoprecipitation of FSP27␤ and the simultaneous decrease of co-immunoprecipitated CideA-MYC with CideA-HA using antibody to HA, indicating that FSP27␤ inhibited the homodimer formation of CideA by binding to CideA. Finally, we investigated whether CideA and FSP27␤ indeed formed a complex in brown adipocytes. We confirmed that FSP27␤ co-immunoprecipitated with CideA using an anti-CideA antibody from detergent extracts of the BAT of wildtype mice (Fig. 7B), which indicates that FSP27␤ binds to CideA in brown adipocytes. However, we failed to confirm the complex formation between CideA and FSP27␤ by the immunoprecipitation experiment using anti-FSP27 antibody (data not shown). We speculate this is because our antibody to FSP27 can not identify FSP27 that forms a complex with CideA.

Discussion
In the present study we confirmed that FSP27␤, a recently identified isoform of FSP27 abundantly expressed in BAT, plays a critical role in the formation of small LD in brown adipocytes by working co-operatively with CideA. FSP27␣ and CideA are enriched at a particular sub-LD location: the LD-LD contact site (15,16). They are both assumed to form a homodimer between contacting LD and mediate LD fusion and growth. Because CideA and FSP27␤ are both localized on the LD surface and form a complex, FSP27␤ is assumed to decrease enlargements in LD by inhibiting the homodimerization of CideA (Fig. 8). This result provides an insight into the mechanisms underlying the formation of small and multilocular LD in BAT.
Adipocytes are required not only to store TAG but also to rapidly hydrolyze TAG in WAT for the supply of FFA to other tissues through the circulation during fasting or for heat production in mitochondria in BAT. Therefore, the unilocular LD form is ideal in WAT because of the efficient storage of TAG and efflux of hydrolyzed FFA from the cell surface to the circulation. FSP27␣ has been suggested to promote the growth of LD by mediating the fusion of contacting LD through the dimerization of FSP27 proteins focally enriched at the contact site between LD (15,18). In addition, FSP27␣ is associated with several proteins including perilipin (18,19) and adipocyte triacylglycerol lipase (20). Because isoproterenol-stimulated lipolysis was previously reported to be impaired in the isolated white adipocytes of FSP27 KO mice (5), FSP27␣ may also contribute to rapid and efficient lipolysis through these proteins in addition to TAG storage. In contrast, in BAT, the formation of mul-tilocular small LD is important for the efficient influx of hydrolyzed FFA to the adjacent mitochondria for ␤ oxidation and subsequent heat production. O 2 consumption stimulated by a ␤3-adrenergic agonist was found to be reduced in isolated brown adipocytes representing the formation of large LD in FSP27 KO mice (5). Furthermore, FSP27␣ may play important roles in rapid lipolysis through adipocyte triacylglycerol lipase (20); therefore, FSP27␤ may also be indispensable for rapid and efficient lipolysis in association with lipases in BAT.
In the present study we demonstrated that CideA and FSP27␤ were mainly expressed in BAT, and the overexpression of CideA promoted the formation of large LD to a similar degree as FSP27␣ in COS cells. A previous study detected multilocular LD in WAT along with the enlargement of LD in BAT from FSP27 KO mice in which FSP27␣ and FSP27␤ were both depleted (5). These phenotypes are explained by the present results showing that FSP27␤ inhibits enlargements in LD in brown adipocytes. In contrast, our results are inconsistent with previous findings demonstrating that the overexpression of FSP27␤ promotes the enlargement of LD in COS7 cells, although the LD size was not quantitated (13). The reason for this difference currently remains unknown. However, it may be related to FSP27␤ being expressed as a fusion protein with GFP in COS cells in the previous study (13). In fact, FSP27␤ expressed as a fusion protein with DSRed in COS cells increased LD size to a small degree in the present study, although the increased level of LD size was much smaller compared with CideA and FSP27␣.
CREBH is a liver-enriched transcription factor localized in the endoplasmic reticulum (ER) membrane and is activated by ER stress or inflammatory stimuli to induce acute-phase hepatic inflammation (21). This transcription factor has also been shown to regulate glucose and lipid metabolism (22)(23)(24). CREBH was recently reported to directly bind to the CREBH response element in the 5Ј-flanking region of the FSP27 gene and to play a critical role in the expression of FSP27␤ in the liver (13). Although FSP27␤ is a major isoform of FSP27 in BAT, CREBH was not expressed in BAT (13). Therefore, the mechanisms responsible for regulating the expression of FSP27␤ in BAT remain unclear. Further studies to elucidate the mechanisms by which the expression of FSP27␤ is regulated in BAT will be important for the development of new therapies to modulate energy metabolism. In conclusion, we herein demonstrated that FSP27␤ and CideA co-operatively form small multilocular LD, which is an advantageous morphology for the transport of FFA to mitochondria adjacent to LD for oxidation in brown adipocytes.

Cell culture and FSP27 knock-out mice
COS cells were cultured in Dulbecco's modified Eagle's medium (DMEM) (Sigma) containing 10% fetal bovine serum (FBS) (Biowest). HB2 cells were prepared from the interscapular BAT of p53 homozygous knock-out mice, as described previously (14). Cells were maintained in DMEM supplemented with 10% FBS (JRH Biosciences), streptomycin (50 g/ml), and penicillin (50 units/ml). At confluence, fresh medium supplemented with 0.5 mM 3-isobutyl-1-methylxanthine and 1 nM dexamethasone was added to cells for the induction of adipocyte differentiation. After 2 days, medium was changed to DMEM supplemented with 50 nM triiodothyronine and insulin (10 g/ml) and was refreshed every 2 days. FSP27 KO mice were generated as previously described (5). Experimental protocols with mice were approved by the Animal Ethics Committee of Kobe University Graduate School of Medicine.

Depletion of FSP27 in HB2 adipocytes
Five days after the onset of induction of differentiation, HB2 adipocytes were washed twice with PBS, detached from the culture dish by exposure to 0.25% trypsin and collagenase (0.5 mg/ml) in PBS, and resuspended in phosphate-buffered saline (PBS). The cells (ϳ3 ϫ 10 6 ) were then mixed with siRNA duplexes and subjected to electroporation with a Bio-Rad Gene Pulser II system at a setting of 0.18 kV and 0.975 microfarads. The FSP27 siRNA was targeted to the mRNA sequence 5Ј-GC-ACAAUCGUGGAGACAGAAGAAUA-3Ј. Immediately after electroporation, the cells were mixed with fresh DMEM supplemented with 10% FBS, and 10 min later they were seeded onto culture plates. They were subjected to assays 2 days after electroporation.

Overexpression of FSP27 and CideA proteins in COS cells and a microscopic analysis
Expression plasmids were introduced into COS cells using X-tremeGENE 9 DNA Transfection Reagent (Roche Applied Science). Before the microscopic analysis, cells were incubated with BODIPY 493/503 (Invitrogen) for the detection of LD at 37°C for 5 min in a cell incubator. Cells were then fixed with PBS containing 4% paraformaldehyde at room temperature, washed with PBS, and exposed to PBS containing 5% bovine serum albumin (BSA). In the immunostaining of FSP27 and CideA, cells were permeabilized with 0.2% Triton X-100 for 5 min and incubated with PBS containing 5% BSA. They were then visualized by indirect immunofluorescence staining with an anti-FSP27 antibody and anti-CideA antibody followed with an Alexa-Fluor555-conjugated goat antibody to rabbit IgG (Invitrogen) or Dylight405-conjugated goat antibody to rabbit IgG (Thermo Scientific). Cells were then examined with a confocal laser-scanning microscope (LSM700, Carl Zeiss). LD sizes in each cell were obtained by measuring the diameter of the largest LD in cells using the measuring instrument installed in the confocal laser-scanning microscope.

Immunoblot and immunoprecipitation analysis
Cell lysates were prepared with lysis buffer containing 25 mM Tris-HCl (pH 7.4), 150 mM NaCl, 1 mM EDTA, 1% Triton X-100, 50 mM NaF, 10 mM sodium pyrophosphate, 1 mM sodium vanadate, and 1 mM phenylmethanesulfonyl fluoride. Adipose tissue homogenates were prepared in the same buffer or CelLytic MT Cell Lysis Reagent (Sigma) using a Teflon homogenizer. In eWAT and BAT, tissue homogenates were subjected to the immunoprecipitation at 4°C for 3 h with antibodies to CideA. Tissue extracts and the resulting immunoprecipitates were subjected to an immunoblot analysis with antibodies to FSP27 and CideA. In COS cells, detergent extracts were subjected to the immunoprecipitation at 4°C for 3 h with antibodies to HA. Total cell and the resulting immunoprecipitates were subjected to an immunoblot analysis with antibodies to MYC, FSP27, and HA. The polyclonal antibody to FSP27 was generated by injecting rabbits with a glutathione S-transferase fusion protein of mouse FSP27 (amino acid residues 45-127) that was expressed in and purified from Escherichia coli. A rabbit antibody to CideA was purchased from Santa Cruz Biotechnology. Mouse antibodies to ␣-tubulin and ␤-actin were purchased from Sigma. A rabbit polyclonal antibody to UCP-1 was purchased from Abcam. Anti-HA antibody produced in rabbit and anti-MYC monoclonal antibody 9E10 was purchased from Sigma and Santa Cruz Biotechnology, respectively.

Histological analysis of mouse BAT
The WAT and BAT of mice were fixed with formalin, embedded in paraffin, sectioned at a thickness of 6 m, and mounted on glass slides using standard procedures. Sections were stained with hematoxylin-eosin. The area and number of LD were measured by fluorescence microscopy (BZ-X710, Keyence).

Statistical analysis
Quantitative data are expressed as the mean Ϯ S.E or Ϯ S.D. The significance of differences between groups was examined with a two-tailed Student's t test. Differences were considered significant at p Ͻ 0.05.
Author contributions-Y. N. and S. N. performed the experiments. S. T. analyzed the data. M. S. contributed the reagents/materials/ analysis tools. W. O. commented extensively on the data and manuscript. Y. T. conceived and designed the experiment and wrote the manuscript. All authors reviewed the results and approved the manuscript.