Maintaining specific cell size, which is important for many organisms, is achieved by coordinating cell growth and cell division. In the budding yeast Saccharomyces cerevisiae, the existence of two cell-size checkpoints is proposed: at the first checkpoint, cell size is monitored before budding at the G1/S transition, and at the second checkpoint, actin depolymerization occurring in the small bud is monitored before the G2/M transition. Morphological analyses have revealed that the small GTPase Rho1p participates in cell-size control at both the G1/S and the G2/M boundaries. One group of rho1 mutants (rho1A) underwent premature entry into mitosis, leading to the birth of abnormally small cells. In another group of rho1 mutants (rho1B), the mother cells failed to reach an appropriate size before budding, and expression of the G1 cyclin Cln2p began at an earlier phase of the cell cycle. Analyses of mutants defective in Rho1p effector proteins indicate that Skn7p, Fks1p and Mpk1p are involved in cell-size control. Thus, Rho1p and its downstream regulatory pathways are involved in controlling cell size in S. cerevisiae.
MAP kinas-activating pathway
To maintain a specific cell size, an organism must coordinate cell growth and cell division. Yeast cells are thought to use cell-size checkpoints to coordinate these two processes (Nurse, 1975; Fantes & Nurse, 1977; Hartwell & Unger, 1977; Johnston, 1977; Rupes, 2002). Cell-size checkpoints prevent cells from passing through cell cycle landmarks until they have reached a critical size. Both external conditions, such as nutrition, and internal sizing mechanisms contribute to cell-size determination (Vanoni, 2005).
The key elements that coordinate cell growth and entry into a new cell cycle are the G1 cyclin-dependent protein kinases. In the yeast Saccharomyces cerevisiae, entry into a new cell cycle depends on the G1 cyclins Cln1p, Cln2p, and Cln3p, and on the cyclin-dependent kinase Cdc28p (Richardson, 1989). Most cyclins are expressed only at specific times during the cell cycle, but only Cln3p is present at relatively constant levels during late G1 and the G1/S transition (Tyers, 1993). Cln3p forms a complex with Cdc28p, and this protein kinase complex activates the transcription factors SBF (Swi4/6 cell-cycle box binding factor) and MBF (MluI cell-cycle box binding factor) (Tyers, 1993; Koch & Nasmyth, 1994). These transcription factors, in turn, elevate the transcription of c. 200 genes in late G1 phase, including CLN1 and CLN2 (Spellman, 1998). Cln1p–Cdc28p and Cln2p–Cdc28p complexes then presumably phosphorylate other substrates such as Sic1p, leading to initiation of the cell cycle. Like CLN3, which was first identified as a mutation that confers a small-cell phenotype (Carter & Sudbery, 1980; Sudbery, 1980), WHI3 was also isolated as a gene involved in cell-size regulation (Nash, 2001). Whi3p has been proposed to act as a cytoplasmic retention device for Cln3p–Cdc28p, thus defining a key cell-size control event in G1 cells (Wang, 2004).
Cell growth and entry into mitosis are coordinated in fission yeast (Sveiczer, 1996). Although similar cell cycle coordination was initially proposed for budding yeast (Harvey & Kellogg, 2003), a recent study indicated that such coordination is obvious only under the actin-stressed condition (McNulty & Lew, 2005), suggesting that budding yeast does not monitor bud size directly. Actin depolymerization is known to affect bud growth and cause Swe1p-dependent phosphorylation of Cdc28p, leading to cell cycle delay (McMillan, 1998; Sia, 1998). By placing small-budded cells under an actin-unstressed condition, McNulty & Lew (2005) succeeded in showing that small bud size itself fails to trigger cell cycle delay. The cell cycle delay caused by actin depolymerization is regulated by the morphogenesis checkpoint. In addition to Swe1p, signaling molecules in the Mpk1p MAP kinase pathway, including Mpk1p, Mkk1p/Mkk2p, Bck1p, Pkc1p, and Rho1p, have been shown to function in the morphogenesis checkpoint (Harrison, 2001).
The Mpk1p MAP kinase pathway is one of the regulatory pathways controlled by Rho1p, a Rho-type GTPase. The GTP- and GDP-bound forms of Rho1p are the active and inactive states, respectively. To date, five proteins have been identified as Rho1p effector proteins. These proteins are Pkc1p, which is implicated in activating the MAP kinase pathway, Fks1p and Fks2p, constituents of 1,3-β-glucan synthase (Drgonova, 1996; Qadota, 1996), Bni1p, which is involved in reorganization of the actin cytoskeleton (Kohno, 1996), Sec3p, which is implicated in vesicular transport (Guo, 2001), and Skn7p, a transcription factor (Alberts, 1998). Temperatur-sensitive rho1 mutants have been characterized and classified into two groups, rho1A (rho1-2 and rho1-5) and rho1B (rho1-3, rho1-4, rho1-10, and rho1-11) (Saka, 2001). Biochemical and cytologic analyses have demonstrated that the rho1A and rho1B mutations cause defects in activation of the Rho1p effectors Pkc1p kinase and 1,3-β-glucan synthase, respectively.
If Swe1p is required for cell cycle coordination, then its upstream signaling molecules in the MAP kinas-activating pathway may function in cell-size control. To understand the role of Rho1p in cell cycle coordination, we analyzed the sizes of several mutant cells defective in Rho1p and its effector proteins. We used the calmorph imag-processing program to automatically characterize each yeast cell using a number of morphological parameters (Ohtani, 2004; Ohya, 2005). As described herein, this detailed analysis of mutant phenotypes reveals a link between Rho1p and cell-size control.
Materials and methods
Media for growth of yeast
Yeast cells were grown in yeast-rich medium [YPD; 1% Bacto yeast extract (Difco, Detroit, MI), 2% Bacto peptone (Difco), and 2% glucose (Wako Fine Chemicals, Osaka, Japan)]. To synchronize the cell cycle, nocodazole was added to the liquid medium at a final concentration of 15 μg mL−1.
The S. cerevisiae strains used in this study are listed in Table 1. The yeast strains are derivatives of the YPH500 strain (Sikorski & Hieter, 1989), and were constructed by standard genetic crosses, transformations, and other genetic procedures (Kaiser, 1994). Yeast transformations were performed using the lithium acetate method (Ito, 1983).
Log-phase cells were fixed for 30 min in growth medium containing 3.7% formaldehyde and 100 mM potassium phosphate buffer (pH 6.5) at 25°C. Cells were sedimented by centrifugation and then incubated in 100 mM potassium phosphate buffer (pH 6.5) containing 4% formaldehyde for 45 min at room temperature. Actin staining was performed by overnight treatment with 20 U mL−1 rhodamine–phalloidin (Molecular Probes, Inc., Eugene, OR) in phosphat-buffered saline (PBS). After the cells were washed three times with PBS, they were mixed with 20 μg mL−1 fluorescein isothiocyanat-concanavalin A (Sigma, St Louis, MO) in P buffer [10 mM sodium phosphate (pH 7.2), 150 mM NaCl] for 5 min to stain mannoprotein on the cell surface. The cells were then mixed with mounting buffer containing 20 μg mL−1 4′-6′-diamidino-2-phenylindole (Sigma) to stain DNA, and 1 μg mL−1p-phenylenediamine (Sigma) to inhibit fading. The specimens were observed under a Zeiss Axiophot2 imaging microscope (× 100 objective; Carl Zeiss, Oberkochen, Germany). Images were captured using a CoolSNAPHQ cooled CCD camera (Roper Scientific Photometrics, Tucson, AZ) interfaced with metamorph software (Universal Imaging Corporation, Downingtown, PA).
The calmorph imag-processing program (Ohtani, 2004) was used to obtain quantitative morphologic data. One pixel is equal to 0.129512 μm. Mother cell and bud volume (fL) were estimated from the measured bud length (l) and width (w) using the formula for an oblate spheroid (v=πlw2/6). The volumes of mother cells and the bud volumes were determined directly using a boxplot.
Cln2p expression analysis
Cells were cultured in 50 mL of YPD overnight, treated with 15 μg mL−1 nocodazole for 4 h, harvested, and resuspended in 50 mL of YPD. After various lengths of time in YPD, samples were collected, washed with distilled water, and placed immediately into a −80°C freezer. Cell pellets were disrupted with glass beads after resuspension in 100 μL of lysis buffer [50 mM Tris-HCl (pH 7.5), 150 mM NaCl, 10% glycerol, 1% Triton X-100, 0.1% sodium dodecyl sulfate (SDS), 50 mM NaF, 1 mM sodium orthovanadate, 50 mM β-glycerophosphate, 5 mM sodium pyrophosphate, 5 mM EDTA, 1 mM phenylmethanesulfonylfluoride, and 25 μg mL−1 each N-tosyl-l-phenylalanyl-chloromethylketone, N-tosyl-l-lysyl-chloromethylketone, leupeptin, pepstatin, antipain, aprotinin, and chymostatin].
For Western blotting analysis, 12-μL aliquots of the disrupted samples were mixed with 4 μL of 4 × sample buffer and resolved by 12% sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE). The separated proteins were transferred to a Hybond-P transfer membrane (Amersham Biosciences, Piscataway, NJ). The blot was developed with an ECL detection kit (Amersham Pharmacia Biotech, Uppsala, Sweden) using an anti-rabbit horseradish peroxidase (HRP) antibody (Cdc28p) and anti-mouse HRP antibody (Cln2p).
Measuring growth rate
Cells were grown in YPD and inoculated in an L-shaped test tube containing 4 mL of YPD. Cell growth was monitored every 15 min by reading the OD at 660 nm with an automated Advantec TVS062CA Bio-Photorecorder (Advantec Toyo, Tokyo, Japan).
The mutations in the Mpk1p MAP kinase pathway cause S. cerevisiae cells to enter mitosis prematurely
To analyze the function of the Mpk1p MAP kinas-activating pathway in coordinating the cell cycle, the bud volumes of mutant S. cerevisiae cells defective in the Mpk1p MAP kinas-activating pathway were analyzed using calmorph (Ohtani, 2004; Ohya, 2005). McNulty & Lew (2005) reported that Δswe1 cells entered mitosis prematurely only when the actin in small-budded cells was depolymerized. Under the unstressed condition, Δswe1 cells enter mitosis with a normal-sized bud (McNulty & Lew, 2005). Likewise, we observed that Δswe1 cells with separated nuclei had normal-sized daughter cells (Fig. 1, Table 2). In contrast, detailed morphological analysis of mutants defective in the upstream Mpk1p MAP kinas-activating pathway showed significantly smaller daughter cells for the Δmkk1Δmkk2, Δbck1 and Δmpk1 mutant cells (Fig. 1, Table 2). In these mutants, entry into mitosis with a small bud occurred because Δmkk1Δmkk2, Δbck1 and Δmpk1 mutants caused actin depolymerization simultaneously (Mazzoni, 1993; Delley & Hall, 1999). The mutants defective in both actin organization and Swe1p activation probably enter mitosis prematurely, before the daughter cells grow sufficiently.
Effects of Δswe1 and Mpk1p MAP kinase pathway deletion mutations on mother cell and bud cell volume after mitosis. (a) Mother cell and bud cell volume distributions for wild-type (open circles) and mutant (filled circles) Saccharomyces cerevisiae cells. The cells were grown in YPD at 25°C, fixed, and tripl-stained. Images were analyzed using the program calmorph. The data for budded cells with one nucleus in the mother cell and one nucleus in the daughter cell are shown. (b) Boxplot of the daughter cell volume at the G2/M transition. The boxplot shows the medians (central horizontal line) with the 25th and 75th percentiles (box). The lower line protruding from the box ends at the 5th percentile, whereas the upper one ends at the 95th percentile. Mutant cells with asterisks have significantly smaller daughter cell volumes than the wild type (Dunnett's test, all P≤0.05). n=200.
Mother cell and daughter cell volumes, and growth rate of the yeast strains
The mean of mother cell volume ± SD at G1/S transition (fL)
The mean of daughter cell volume ± SD at G2/M transition (fL)
Doubling time (h)
54.40 ± 6.63
23.14 ± 0.79
50.98 ± 2.02
24.74 ± 1.29
54.31 ± 5.57
25.27 ± 3.38
50.57 ± 10.54
21.77 ± 2.79
47.17 ± 2.82
18.56 ± 2.03
56.44 ± 7.75
18.40 ± 2.38
49.84 ± 10.82
18.35 ± 1.54
44.38 ± 4.94
18.63 ± 2.47
44.21 ± 2.35
19.65 ± 1.64
37.33 ± 2.87
20.86 ± 1.10
41.05 ± 3.17
22.56 ± 1.62
36.70 ± 3.30
20.56 ± 0.86
40.68 ± 3.17
19.84 ± 0.47
27.43 ± 1.09
13.92 ± 0.79
27.19 ± 1.49
14.74 ± 0.71
The means of mother cell volume ± SD and daughter cell volume ± SD were derived from five independent experiments, at G1/S transition and G2/M transition, respectively.
Doubling times (h) were measured by bio-photorecorder (TAITEC) at 30°C.
↵* Significantly smaller mother or daughter cell volume than the that of the wild type (Dunnet's tests, all P-values less than or equal to 0.05).
rho1A mutant cells prematurely enter mitosis
As Rho1p is essential for Mpk1p MAP kinase activity, we next examined whether Rho1p is involved in bud-size control at the G2/M boundary. Temperatur-sensitive rho1 mutants have been classified into two groups, rho1A and rho1B. We found that the rho1A (rho1-2 and rho1-5) mutant cells with divided nuclei had daughter cells that were smaller than the wild-type daughter cells (Fig. 2). Consequently, the volume of the rho1A (rho1-2 and rho1-5) mutant daughter cells were smaller than those of the wild-type cells (Fig. 2, Table 2). As the rho1A mutation specifically results in a defective Mpk1p MAP kinase pathway, the decreased activity of the MAP kinase pathway in rho1A cells probably explains the premature entry of rho1A cells into mitosis. In contrast, the volumes of the rho1B (rho1-3 and rho1-4) mutant daughter cells were similar to those of the wild-type cells (Fig. 2, Table 2).
Effect of rho1 mutations on mother cell and daughter cell volume after mitosis. (a) Mother cell and bud cell volume distributions for wild-type (open circles) and mutant (filled circles) cells. The cells were grown in YPD at 25°C, fixed, and tripl-stained. Images were analyzed using calmorph. The data for budded cells with one nucleus in the mother cell and one nucleus in the daughter cell are shown. (b) Boxplot of the daughter cell volume at the G2/M transition. Mutant cells with asterisks have significantly smaller daughter cell volumes than the wild type (Dunnett's test, all P≤0.05). n=200.
rho1B mutant cells bud at an early stage
Next, we analyzed the volumes of the mother cells of singl-nucleus budded cells. Consistent with a previous report that whi3 mutant cells are small (Nash, 2001), our Δwhi3 mutant cells had smaller mother cells (Fig. 3, Table 2). In addition, rho1B (rho1-3 and rho1-4) mutant cells had smaller mother cells, whereas rho1A (rho1-2 and rho1-5) mutant cells had mother cells of normal volume (Fig. 3, Table 2). These data suggest that the cell-size checkpoint at the G1/S transition of rho1-3 and rho1-4 mutants is defective. rho1-4 mutants grew more slowly than rho1-3 mutants, but no correlation was observed between the growth rate and mother cell volume (Table 2). As yeast cells growing slowly on a poor medium generally pass START at a smaller size than fast-growing cells on a rich medium, the slow-growth phenotype of rho1-4 cells was not due to nutrient limitation.
Effect of rho1 mutations on the volume of daughter cells and singly nucleated mother cells. (a) Mother cell and bud cell volume distributions for wild-type (open circles) and mutant (filled circles) cells. The cells were grown in YPD at 25°C, fixed, and tripl-stained. Images were analyzed using calmorph. The data for budded cells with one nucleus in the mother cell are shown. (b) Boxplot of mother cell volume distributions. The morphological data used in (a) were analyzed using a boxplot to obtain the mother cell volumes of the wild-type and mutant cells. Mutant cells with asterisks have significantly smaller mother cells than the wild type (Dunnett's test, all P≤0.05). n=200.
To further investigate the role of Rho1p in controlling cell size at the G1/S transition, we constructed a rho1-3Δwhi3 double mutant. The WHI3 deletion and rho1-3 mutation did not exhibit an additive effect in decreasing mother cell volume at the G1/S transition; the cell-size distribution of the double mutants was similar to that of Δwhi3 cells (Table 2). These results suggest that rho1-3 mutant cells have a cell-siz-control defect that occurs in the same pathway as that of Δwhi3 cells.
Morphological effects of mutations in Rho1p effector proteins
To investigate which Rho1p effector proteins play a role in cell cycle coordination, we examined the volumes of mutant cells with various Rho1p effector deletions. As fks1-1125, a temperatur-sensitive mutation in 1,3-β-glucan synthase, causes a significant decrease in 1,3-β-glucan synthase activity, even at permissive temperatures (Dijkgraaf, 2002), we analyzed the effects of fks1-1125Δfks2, Δskn7, Δbni1, Δbck1, Δmkk1Δmkk2 and Δmpk1 mutations on cell size at 30°C. We found that the daughter cell volumes of the Δmpk1, Δbck1 and Δmkk1Δmkk2 mutants were significantly smaller, whereas those of the other Rho1-effector protein deletion mutants were almost the same as those of the wild type (Fig. 4b, Table 2). An examination of the volume distributions of budded cells showed that fks1-1125, Δfks2 and Δskn7 mother cells were smaller than wild-type mother cells (Fig. 4a and c). Notably, the small-cell phenotypes of fks1-1125, Δfks2 and rho1B mutants were similar, consistent with the fact that rho1B mutants are defective in activation of 1,3-β-glucan synthase (Saka, 2001).
Effect of Rho1p effector mutations on the volume of daughter cells and singly nucleated mother cells. (a) Mother cell and bud cell volume distributions of wild-type (open circles) and mutant (filled circles) cells. The cells were grown in YPD at 25°C, fixed, and tripl-stained. Images were analyzed using calmorph. The data for budded cells with one nucleus in the mother cell are shown. (b) Boxplot of the mother cell volume distributions. The morphological data shown in (a) were analyzed using a boxplot to obtain the mother cell volumes of the wild-type and mutant cells. Mutant cells with asterisks have significantly smaller mother cells than the wild type (Dunnett's test, all P≤0.05). n=200. (c) Boxplot of the daughter cell volume at the G2/M transition. Mutant cells with asterisks have significantly smaller daughter cell volumes than the wild type (Dunnett's test, all P≤0.05). n=200.
Cln2p expression in rho1-3 mutant cells
As our results showed that rho1-3 mutant cells form buds before reaching critical cell volume, we further examined this phenomenon by analyzing the timing of Cln2p expression in wild-type and mutant cells using Western blotting (Fig. 5). G2/M cells were synchronized using nocodazole, released into fresh YPD medium, and analyzed for Cln2p expression. Under our experimental conditions, as nocodazol-arrested cells have a dumb-bell phenotype of large budded cells, it is a concern that the volume of the bud exceeds the threshold value of the bud volume at G1/S phase. However, the mother cell volume (51.28±7.77 fL) and daughter cell volume (24.83±2.12 fL) at G2/M phase were under the threshold volume (54.4±6.63 fL) of the mother cell at G1/S phase. In wild-type cells, Cln2p expression appeared after 90 min, coincident with the onset of bud emergence (data not shown). In contrast, Cln2p expression was evident at 70 min in Δwhi3 cells, consistent with the earlier onset of bud emergence (data not shown). Cln2p expression in rho1-3 cells was slightly accelerated, as it was also in the fks1-1125 and Δskn7 cells (Fig. 5). These results provide further evidence that Rho1p and its Rho1p effector proteins are involved in cell-size control at the G1/S transition.
Western blot analysis of Cln2p levels. Samples were taken from synchronized cells arrested by nocodazole (0 min) or released from arrest for the indicated lengths of time (20–120 min). The Cdc28p band was used to control for variances in gel loading.
The results of this study revealed the relationship between Rho1p and cell-size control in S. cerevisiae. Our data suggest that Rho1p is involved in cell-size control through two distinct pathways: one monitors mother cell volume before the onset of bud emergence, and the other regulates daughter bud volume before mitosis. Interestingly, rho1A mutants (rho1-2 and rho1-5) have phenotypes similar to those of theΔmpk1 mutants, and the phenotypes of the rho1B mutant (rho1-3 and rho1-4) cells resemble that of the Δwhi3 mutant, possibly because the different rho1 alleles have distinct functional defects in the activation of downstream pathways.
Our data indicated that the volumes of Δswe1 cells are similar to those of the wild type at mitosis, whereas mutants defective in the upstream Mpk1p MAP kinas-activating pathway force cells to enter mitosis prematurely. As loss of the Mpk1p MAP kinase activity affects actin polymerization, the actin-stressed condition generated in mutants defective in the Mpk1p MAP kinas-activating pathway induces the morphological checkpoint. Although Δswe1 cells enter mitosis with a small buds only under the stressed condition (McNulty & Lew, 2005), mutants defective in the Mpk1p MAP kinas-activating pathway enter mitosis with a small bud even under normal growth conditions. Although it is not clear whether the Mpk1p MAP kinas-activating pathway monitors the actual bud volume, our data showed that the Mpk1p MAP kinas-activating pathway is required for the normal increase in bud volume before G2/M (Fig. 6). rho1A mutants (rho1-2 and rho1-5) have phenotypes similar to that of Δmpk1 cells, suggesting that Rho1p is also important for bud maturation before mitosis (Fig. 6).
A model for the role of Rho1p in cell-size control.
Our findings that rho1-3 mutants have a reduced cell volume and also express Cln2p earlier than wild-type cells suggest that the rho1B mutation results in a defect in the pathway that monitors critical cell volume at the G1/S transition (Figs 3 and 5). A functional interaction between Rho1p and Whi3p was suggested by our observation that the effects of the rho1B and Δwhi3 mutations on mother cell volume were not additive in the rho1BΔwhi3 mutant (Table 2). The reduction in the volume of rho1-3 mutant cells was not as severe as that of Δwhi3 cells, but this difference probably occurred because the experiments were carried out at permissive temperatures, resulting in only mild defects. Whi3p, a negative G1 regulator of Cln3p, is associated with Cdc28p in vivo, restricting its cellular distribution to the cytoplasm (Wang, 2004); however, at present, we do not have any evidence showing that Rho1p directly regulates this function of Whi3p. In addition, Whi3p and Rho1p localize differently in the cell; Whi3p interacts with Cdc28p and colocalizes with CLN3 mRNA to cytoplasmic foci, whereas Rho1p localizes at the plasma membrane (Yamochi, 1994; Gari, 2001). Recently, it was proposed that G1 delay of daughter cells is not related to cell size but is specified by asymmetric distribution of Ace2p (Laabs, 2003). Deletion of Ace2p leads to daughter cells that proceed through G1 at the same rate as mother cells, and thus decreases the average size of mother cells at late G1 phase. Therefore, it is possible that asymmetric distribution of Ace2p determines the daughter cell size by regulating G1 cyclin expression. It would be worthwhile to independently analyze the volume of unbudded cells derived from the daughter cells and mother cells. However, the current version of the calmorph imag-processing program cannot follow the formation of mother–daughter pairs, so this would require upgrading of calmorph.
These results are consistent with a model in which Rho1p senses the volume of the cell and transmits a signal to Whi3p, Ace2p or another cell-siz-control protein around the nucleus via an unknown protein or proteins. Skn7p and Fks1p are possible candidates for components of the downstream pathway, as their mutant phenotypes were similar to the rho1-3 phenotype (Fig. 4c). Skn7p functions in several stress responses, including responses to cell integrity breaches, oxidative stress, and heat shock (Hohman, 2002), although its role in such a variety of processes is not completely understood (Williams & Cyert, 2001). Skn7p has also been reported to associate with Mbp1p, the DNA-binding component of the G1-transcription factor DSC1/MBF (Bouquin, 1999). Therefore, Skn7p may participate in cell-size control at the G1/S transition downstream of Rho1p in several ways (Fig. 6). Fks1p is a component of the 1,3-β-glucan synthase complex (Dijkgraaf, 2002). One explanation for the effects of the rho1 alleles is that they affect cell wall synthesis via Fks1p; cells with thinner cell walls will appear smaller, even if the mass of the cytoplasm remains constant. However, this seems unlikely, as electron microscopic observations revealed that the cell wall of the fks1-1125 mutants is not thinner than that of the wild-type strain (A. Hirata and Y. Ohya, unpublished results).
With the imag-processing program calmorph, we were able to perform a precise statistical and quantitative analysis of mother cell and bud cell volumes for wild-type and mutant cells. Further research based on this phenotypic data should reveal the other factors involved in cell-size regulation, increasing our understanding of the molecular network that coordinates cell growth and division.
A preliminary analysis of the data was performed by A. Saka. We thank S. Ohnuki for assistance with the statistical analysis. This work was supported by a Grant for Scientific Research (#16026205) from the Ministry of Education, Culture, Sports, Science and Technology of Japan, and by the Institute for Bioinformatics and Research and Development of the Japan Science and Technology Corporation.