Aberrant TET1 Methylation Closely Associated with CpG Island Methylator Phenotype in Colorectal Cancer

Colorectal cancer is one of the most frequent malignancies worldwide and is associate with accumulations of genetic and epigenetic alterations (1). Studies of colorectal cancers have indicated the existence of a CpG island methylator phenotype (CIMP) in a subset of cases, in which a high rate of aberrant DNA methylation was present and characteristic clinical features were exhibited, such as high rates of BRAF or KRAS mutations, specific histology (mucinous, poorly differentiated), proximal location, serrated polyp pathway, and better prognosis (2). In addition, recent studies suggested that CIMP is an early, possibly tumor-initiating event (3). This accumulated evidence indicates that CIMP-positive (CIMP-P) colorectal cancers may have distinct clinicopathological manifestations and develop through a unique pathway (4).

Although many research studies have demonstrated the existence of CIMP in colorectal cancers, a key missing piece of information is the underlying cause of CIMP. DNA methylation is catalyzed by DNA methyltransferase (DNMT) family members, DNMT1, DNMT3A, and DNMT3B. The relationship between expression levels of DNMTs and CIMP status in colorectal cancers has been extensively studied (5–7). However, conclusive observations have not been demonstrated so far, suggesting that accumulation of DNA methylation is not simply explained by over-activation of DNMTs.

Recently, members of the ten-eleven translocation (TET) family, TET1, TET2, and TET3, which catalyze the conversion of 5-methylcytosine (5 mC) into 5-hydroxymethylcytosine (5 hmC) were identified (8). As this conversion is a primary mode of DNA demethylation, impaired TET activity may affect DNA hypermethylation.

Somatic mutations of isocitrate dehydrogenase 1 (IDH1) produces 2-hydroxyglutarate (2-HG), which inhibits the catalytic activity of TET families and consequently induces accumulation of aberrant DNA methylation (9, 10). Indeed, studies demonstrated that IDH1 mutations in glioma may be sufficient to establish CIMP in glioma (10, 11). Another study of acute myeloid leukemia (AML) also revealed that IDH1 or TET2 mutations induce DNA hypermethylation (12). These studies indicate that the demethylation process mediated by TET may have a causative role in various types of CIMP-P tumors. However, inactivation of TET family members and the corresponding clinical impact of such alterations in CIMP-P colorectal cancers have not been well studied.

In this study, we found that TET1 is frequently silenced by DNA methylation in CIMP-P colorectal cancers. TET1 methylation appears to be an early event, even frequently found in precancerous lesions, and is associated with a global effect on DNA methylation status, as well as clinical outcome of colorectal cancers. Our data shed light on the clinical impact of TET1 methylation during colorectal cancer tumorigenesis.

Six colorectal cancer cell lines, SW480, SW48, RKO, HT29, LOVO, and LS174T as well as a breast cancer cell line, MCF7 were purchased from the American Type Culture Collection. Although these cell lines were not authenticated, they were relatively young passage cells after being obtained. SW480, RKO, HT29, LOVO, LS174T, and MCF7 cells were maintained in D-MEM medium (Life Technologies). SW48 was maintained in L-15 medium (Life Technologies). Colorectal cancer cell lines were classified as CIMP-P (SW48, RKO, HT29) or CIMP-N (SW480, LOVO, LS174T), as described previously (13). All cell lines were supplemented with 10% fetal bovine serum (Life Technologies) and antibiotic–antimycotic reagent (Life Technologies) at 37°C in a humidified incubator with 5% CO₂. SW48, RKO, and HT29 cells were treated with 1 μmol/L 5-aza-2′-deoxycytidine (DAC; Sigma-Aldrich) as described previously for 72 hours, with the drug and medium replaced every 24 hours (14).

Samples of colon polyp (n = 64), colorectal cancers (n = 168) and their corresponding normal colonic mucosal tissues (n = 63) were collected from individuals who underwent surgical resection at the Aichi Cancer Center Central Hospital and Sapporo University Hospital, Japan, after appropriate institutional review board approval was received and written informed consent had been obtained. The background information of the patients and associated colorectal cancer tumor specimens used are shown is in Supplementary Table S1. Colon polyps that were assessed consisted of 30 sessile serrated adenoma/polyps (SSA/P), 18 traditional serrated adenomas (TSA), 6 hyperplastic polyps (HP), and 10 tubular or tubulovillous adenomas. Among these, 11 out of the 30 SSA/P and one of the 10 adenomas were mixed lesions containing both adenoma and adenocarcinoma components in the same polyp.

Total RNA was isolated using TRizol (Life Technologies), with 2 μg used for reverse transcription with a SuperScript VILO cDNA Synthesis Kit (Life Technologies). TaqMan PCR and SYBR Green qPCR were carried out for the target genes in duplicate. Oligonucleotide primers used for TaqMan PCR assays (Applied Biosystems) are as follows, GAPDH (AB Assay ID Hs00266705_g1) and proliferating cell nuclear antigen (PCNA, Hs00952870_g1). Primer sets for SYBR Green assays are shown in Supplementary Table S2.

Cell lysates were extracted from SW480, SW48, RKO, HT29, LOVO, LS174T, and MCF7 cells. Cell lysates were also extracted from 16 clinical colorectal cancer samples. Fifty micrograms of each lysate was run on SDS/PAGE gels, transferred to nitrocellulose membranes, and incubated with the following antibodies: rabbit polyclonal anti-TET1 (GTX 124207; GeneTex), rabbit polyclonal anti-TET2 (ab94580; Abcam), rabbit polyclonal anti-TET3 (GTX 121453; GeneTex), rabbit monoclonal anti-MLH1 (ab92312; Abcam), and rabbit polyclonal anti-β-actin (4967; Cell Signaling Technology).

We performed bisulfite treatment on extracted genomic DNA as described previously (15). The DNA methylation levels were measured using pyrosequencing technology (Pyrosequencing AB). Primer sequences and PCR conditions are shown in Supplementary Table S2. CIMP status of 168 colorectal cancers and 64 polyps was evaluated by five classical CIMP markers (hMLH1, MINT1, MINT2, MINT31, and p16). DNA methylation status (positive or negative) of CIMP markers was determined as described previously (16). Samples with simultaneous methylation of at least two of the five classical CIMP markers were considered CIMP-P (17, 18).

Global analysis using methylated CpG island amplification-microarray technology was carried out using DNA from 22 colorectal cancer samples (five CIMP-P with TET1 methylation, six CIMP-P without TET1 methylation, and 11 CIMP-N colorectal cancers) as described previously (14, 16, 19).

We performed target resequencing of 16 CIMP-P colorectal cancers using the HaloPlex Cancer Research Panel, which consists of 47 cancer-associated genes (Agilent Technologies), according to the manufacturer's protocol. Sixteen samples were randomly selected from CIMP-P colorectal cancers. Briefly, 900 ng of enzymatically digested DNA was hybridized to specific probes for each gene of interest. Hybridized molecules were captured and amplified using indexed primers. Following this, target libraries were purified using an AMPure XP Kit (Beckman Coulter) and sequenced via use of a high-throughput sequencing instrument (MiSeq, Illumina). Data were analyzed using CLC genomic Workbench v6.5.1 (CLC Inc.).

In addition, mutations in KRAS (codons 12 and 13) or BRAF (codon 600) were determined in all colorectal cancer cases (n = 168) by the pyrosequencing analysis, as previously described (16). Mutations in TET1 and IDH1 were examined in CIMP-P colorectal cancers (n = 13 and 31, respectively) by conventional or high-throughput sequencing analysis. The PCR primer sequences used in both contexts are listed in Supplementary Table S2.

We obtained the genetic alterations, DNA methylation, and clinicopathological data for 593 colorectal cancers, which included the DNA methylation status of TET1, hMLH1, and CIMP markers, from TCGA web site (http://tcga-data.nci.nih.gov/tcga/tcgaHome2.jsp). Regarding the methylation analysis using the Illumina Infinium Human DNA Methylation 27 and 450 platforms, a β-value > 0.2 was considered as methylation-positive. As one of the classical five CIMP markers (MINT31) was not present in either of these Illumina Infinium platforms, we used an alternative set of five CIMP markers, namely, CACNA1G (Probe ID on the Infinium HumanMethylation 27 or 450 array, cg11262815), IGF2 (cg13756879), NEUROG1 (cg11248413), RUNX3 (cg22737001), and SOCS1 (cg06220235), as reported previously (20, 21). Although CIMP status was determined by a different collection of markers in the TCGA data set, the clinical features of these TCGA CIMP-P colorectal cancers were consistent with CIMP-P colorectal cancers in our analyses as identified by the classical five CIMP markers (Table 1).

All statistical analyses were performed using GraphPad Prism 5 (Graphpad Software) and JMP statistical software version 5.1 (SAS Institute).

The Fisher's exact test was used to determine nonrandom associations between two categorical variables. Kruskal–Wallis analysis was used to evaluate the extent of differences among more than three groups. All reported P values are two sided, with P < 0.05 taken as statistically significant.

First, expression of TET family members (TET1, TET2, and TET3) was examined by qRT-PCR and Western blot analyses in six colorectal cancer cell lines, SW48 (CIMP-P), RKO (CIMP-P), HT29 (CIMP-P), LS174T (CIMP-N), SW480 (CIMP-N), and LOVO (CIMP-N) along with a breast cancer cell line, MCF7, which was used as a positive control for TET gene expression (ref. 22; Fig. 1A). The CIMP status of each colorectal cancer cell line was determined previously (13).

Neither mRNA nor protein expression of TET2 was detected in any of the colorectal cancer cell lines examined, in contrast to MCF7 cells that highly expressed TET2. TET3 was substantially expressed in all colorectal cancer cell lines regardless of CIMP status. Intriguingly, TET1 expression negatively correlated with CIMP status; indeed, significantly lower expression of TET1 was observed in CIMP-P cell lines (SW48, RKO, HT29) than CIMP-N cell lines (LS174T, SW480, LOVO) at both mRNA and protein levels. These results suggest a possible association between TET1 and CIMP status in colorectal cancers.

Next, in order to address the mechanism of TET1 silencing in colorectal cancer cell lines, we examined the DNA methylation status of the TET1 gene by bisulfite pyrosequencing analysis (Fig. 1B and Supplementary Fig. S1A). Three colorectal cancer cell lines without TET1 expression (SW48, RKO, HT29) showed a high level of DNA methylation, whereas the other three cell lines with TET1 expression (LS174T, SW480, and LOVO) showed low level of DNA methylation (Fig. 1B). Silencing of TET1 expression was restored by DAC treatment in SW48, RKO, and HT29 cell lines (Fig. 1C). These results indicate that silencing of TET1 is closely associated with DNA methylation in colorectal cancer cell lines.

We next examined TET1 expression by qRT-PCR in paired cancerous and corresponding normal tissues from colorectal cancer cases (n = 63). TET1 expression was significantly downregulated in cancerous tissues compared to normal tissues (P < 0.0001; Supplementary Fig. S1B). Among them, 51 of 63 colorectal cancers (81%) showed lower levels of TET1 expression, 11 of 63 colorectal cancers (17%) showed almost equal levels, and 1 of 63 colorectal cancers (2%) showed higher levels, respectively (Supplementary Fig. S1C).

Intriguingly, among 119 colorectal cancer cases, including aforementioned 63 paired colorectal cancer cases, TET1 expression was significantly lower in CIMP-P (n = 28) than CIMP-N (n = 91) colorectal cancers (P = 0.0063) as was found in cell line analysis (Figs. 1A and 2A). By contrast, no obvious differences in TET1 expression were observed between normal tissues from CIMP-P cases (n = 16) and CIMP-N cases (n = 47) (P = 0.66; Fig. 2A).

The inverse relationship between TET1 methylation and protein expression was observed in some colorectal cancers (Supplementary Fig. S1D). We further investigated the relationship between TET1 methylation and expression levels in colorectal cancers by bisulfite pyrosequencing analysis and qRT-PCR, respectively. A number of colorectal cancers (11 of 119 colorectal cancers, 9%) showed an increased level of DNA methylation (>8%) and significant reduction of TET1 expression (Fig. 2B). Therefore, we defined DNA methylation levels greater than 8% as TET1 DNA methylation-positive hereafter.

We further added 49 colorectal cancers from Sapporo University Hospital in our DNA methylation analysis. Among the 168 colorectal cancers, 23 of 55 (42%) CIMP-P colorectal cancers showed TET1 methylation, whereas only 2 of 113 (2%) CIMP-N colorectal cancers showed TET1 methylation (P < 0.0001; Fig. 2C). In other words, 23 of the 25 TET1 methylated colorectal cancers were CIMP-P (92%). Thus, TET1 silencing by DNA methylation was closely associated with CIMP status during colorectal cancer tumorigenesis.

We next endeavored to clarify when TET1 methylation emerges and whether TET1 methylation is a cause of CIMP or a downstream target of CIMP-related hypermethylation during colorectal cancer tumorigenesis. Sixty-four colon polyps (a precancerous condition), consisting of 30 SSA/P, 18 TSA, 6 HP, and 10 tubular or tubulovillous adenoma, were examined. TET1 methylation was significantly higher in CIMP-P polyps than CIMP-N polyps [16/40 (40%) vs. 2/24 (8%), P = 0.0088; Fig. 3A]. Thus, the majority of TET1 methylated polyps (16/18, 89%) were CIMP-P as was found in colorectal cancers.

We found that TET1 methylation was highly frequent in CIMP-P tumors with BRAF mutation (BRAF^V600E, P = 0.005; Supplementary Table S3). The existence of distinct classes of CIMP tumors has been shown recently. A subset of CIMP-P colon polyps, as well as colorectal cancers, has been demonstrated to harbor BRAF mutation and exhibit characteristic clinical features, such as more extensive methylation, better prognosis, proximal location, and hMLH1 methylation, called CIMP1 or CIMP-high (CIMP-H; refs. 18, 23, and 24). We examined TET1 methylation in CIMP1/CIMP-H colon polyps and colorectal cancers, as defined by these criteria. A trend toward higher frequency of TET1 methylation was found in colorectal cancers (15/23, 65%) than polyps (14/32, 44%), although this difference was not statistically significant (Fig. 3B, P = 0.17). Intriguingly, DNA methylation of hMLH1, which is another characteristic target of CIMP1/CIMP-H tumors, was more cancer-specific [polyps, 3/32 (9%); colorectal cancers, 16/23 (70%); Fig. 3C, P = 0.001]. Inverse relationship between hMLH1 methylation and protein expression was also found in some colorectal cancers (Supplementary Fig. S1D).

Colorectal cancers are generally evolved from a monoclonal cell population. Therefore, molecular dissection of precancerous (adenoma) and cancerous (adenocarcinoma) components within the same tumor may help to understand the onset timing of CIMP during tumorigenesis (25). We analyzed mixed lesions containing both precancerous and cancerous components in CIMP-P tumors and found that CIMP status and/or BRAF mutation were frequently observed (10/12, 83% and 9/12, 75%, respectively) in precancerous components, whereas TET1 methylation was much less common (2/12, 17%, P = 0.002; Fig. 3D). In addition, TET1 methylation was more frequent in cancerous than precancerous components.

Next, we analyzed the relationship between TET1 methylation and other CIMP markers. Levels and/or frequencies of DNA methylation in four CIMP-marker genes (MINT1, MINT2, MINT31, and p16) was significantly higher in CIMP-P colorectal cancers with TET1 methylation than those without TET1 methylation (Fig. 4A and B and Supplementary Table S3). Of interest, TET1 methylation status was not correlated with hMLH1 methylation status in CIMP-P colorectal cancers (Fig. 4A and B and Supplementary Table S3). In addition, around half (12/23, 52%) of CIMP-P colorectal cancers with TET1 methylation did not show hMLH1 methylation (Fig. 4D).

Next, genome-wide CpG methylation status was compared between colorectal cancers with and without TET methylation using MCAM technology (14, 16). We assessed 22 colorectal cancers in this context: five CIMP-P colorectal cancers with TET1 methylation, six CIMP-P tumors without TET1 methylation and 11 CIMP-N tumors without TET1 methylation. The greatest number of DNA methylated genes were observed in CIMP-P tumors with TET1 methylation compared with the other two groups (P = 0.007; Fig. 4C).

We further analyzed the relationship between genome-wide DNA methylation status, TET1 methylation, and hMLH1 methylation in a more extensive cohort (CIMP-P, n = 229) from a TCGA data set (Supplementary Fig. S2B). CIMP-P colorectal cancers with either TET1 or hMLH1 methylation showed a greater number of DNA methylated genes than those with neither TET1 nor hMLH1 methylation. Notably, colorectal cancers with both TET1 and hMLH1 methylation showed the greatest number of DNA methylated genes compared to other subgroups of colorectal cancers (P < 0.0001).

Studies of leukemia and brain tumors showed aberrant accumulation of DNA methylation was caused by genetic alterations, such as IDH mutation (10, 12, 26). We first examined mutations in TET1 in CIMP-P colorectal cancers without TET1 methylation (n = 13) and found no cancer-specific mutation in the TET1 gene. We next examined genetic alterations of 47 cancer-associated genes, including IDH1, IDH2, BRAF, KRAS, and TP53 and other cancer-associated genes, in 16 CIMP-P colorectal cancers. Concordant with previous reports, mutations in BRAF and KRAS were frequently found in CIMP-P colorectal cancers [five cases (31%) and six cases (38%), respectively; refs. 18, 27]. By contrast, IDH1 and IDH2 mutations were rare (1 case, 6%; Supplementary Fig. S2A). In addition, we examined data relating to 224 colorectal cancers from TCGA data and found TET1, IDH1, and IDH2 mutations in 10 (4%), 3 (0.5%), and 7 (1%) colorectal cancers, respectively, indicating that TET1 and IDH mutations are rare events in colorectal cancer (1).

Finally, relationships between TET1 methylation and clinicopathological features were studied (Table 1). Among 168 colorectal cancers, we found that TET1 methylation was more frequent in elderly patients (P = 0.029), or tumors that were proximally located (P = 0.0001), poorly or mucinous differentiated (P = 0.005), CIMP-P (P < 0.0001), and BRAF mutated (P < 0.0001).

We also extended the analysis of relationship between TET1 methylation and clinicopathological features in additional colorectal cancer cohorts using the TCGA data set (Table 1; ref. 1). Although CIMP status was determined using different markers (CACNA1G, IGF2, NEUROG1, RUNX3, and SOCS1) in the TCGA data set (20), CIMP status was also closely associated with TET1 methylation status. Among 593 colorectal cancers, 75 of 229 CIMP-P colorectal cancers (33%) showed TET1 methylation, whereas 4 of 364 (1%) CIMP-N colorectal cancers showed TET1 methylation (P < 0.001; Table 1). In addition, we found that the clinical features of colorectal cancers with TET1 methylation within the TCGA data set were quite consistent with our data (Table 1).

Next, we further compared the clinical features between CIMP-P colorectal cancers with and without TET1 methylation in the TCGA data set, which contains TNM (for tumors/nodes/metastases) staging information (n = 201). Intriguingly, patients that had colorectal cancers with TET1 methylation displayed less pathological metastatic lesions than those without TET1 methylation (Supplementary Table S4). In addition, among this cohort, TET1 methylation and hMLH1 methylation plus TET1 methylation were the only factors that influenced the incidence of metastasis (P = 0.045 and 0.002, respectively; Table 2).

Recent accumulating evidence has demonstrated that dysregulation of epigenetic modifiers is frequently involved in tumor formation (28). In this study, we focused on the demethylation enzyme TET1 to elucidate the relationships between induction of aberrant DNA methylation and dysregulation of the associated gene in colorectal cancers.

The TET1 gene was first discovered as a translocation partner of the mixed lineage leukemia (MLL) gene in acute myeloid leukemia (29). Recent studies revealed that TET1 protein has a potential mechanism related to DNA demethylation and may play an important role in transcriptional regulation of genes involved in basic housekeeping processes and those regulating differentiation and development (8, 26, 30). Dysregulation of the TET1 gene has been found in many types of cancers and appears to be associated oncogenic processes in solid tumor tissues, such as colon, prostate, breast, and hepatocellular cancers (22, 31–33). Suppression of TET1 promotes tumor cell invasion and correlated with poor survival outcomes in breast cancer patients (22). Even given its putative role in DNA demethylation, no prior study had so far elucidated relationships between TET1 and accumulation of aberrant DNA methylation in colorectal cancers, especially in CIMP-related cancers.

In this study, we initially found that DNA methylation of the TET1 gene is associated with loss of its expression in colorectal cancers. The DNA methylation inhibitor DAC restores TET1 expression in colorectal cancer cell lines, indicating that the DNA hypermethylation plays a major role in silencing of this gene in colorectal cancer. Notably, TET1 methylation is tumor-specific and tightly associated with CIMP-P colorectal cancers, especially those carrying BRAF mutation.

The existence of multiple parallel pathways of tumorigenesis (at least three) has been indicated via DNA methylation profiling in colorectal cancers, namely two CIMP-P options (either CIMP1/CIMP-H and CIMP2/CIMP-L) or CIMP-N (18, 23, 24). These distinct oncogenic pathways raise several clinical implications in terms of therapeutic strategy. Among these, the CIMP1/CIMP-H pathway is characterized by serrated precursor lesions, good prognosis, extensive DNA hypermethylation, BRAF mutation, and microsatellite instability phenotype with hMLH1 methylation. Of interest, we found around half of CIMP-P colorectal cancers with BRAF mutation showed TET1 methylation but not hMLH1 methylation and that TET1 methylation was significantly associated with BRAF mutation but not with hMLH1 methylation in the CIMP-P colorectal cancers examined. Further, we found that TET1 methylation was significantly associated with genome-wide DNA hypermethylation status in CIMP-P colorectal cancers (Fig. 4C). Additionally, colorectal cancers with TET1 methylation plus hMLH1 methylation were the most extensively methylated (Supplementary Fig. S2B). These data suggest that TET1 methylation may contribute to the establishment of unique pathway in CIMP-mediated tumorigenesis, which may incidentally led to hMLH1 methylation. Given the molecular function of TET1, it is reasonable that inactivation of the TET1 gene may induce the accumulation of DNA methylation during colorectal cancer tumorigenesis, as has been reported in glioma and AML studies (10–12).

Interestingly, a high frequency of TET1 methylation was found even in precancerous tissue indicating that TET1 methylation is an early event and may affect global DNA methylation status during tumorigenesis in a subset of tumors, although it remains unclear what triggers the transition to cancer in these polyps. In addition, the occurrence of TET1 methylation in precancerous tissue enabled us to perform molecular dissection in mixed lesions containing both precancerous and cancerous components in CIMP-P tumors. A higher frequency of CIMP positivity and BRAF mutation was found compared to TET1 methylation, as well as hMLH1 methylation, in precancerous components: the frequency of both TET1 methylation and hMLH1 methylation was increased in cancerous components, although hMLH1 methylation was more cancer specific. Taken together, these data indicate that TET1 methylation occurs after acquiring BRAF mutation and CIMP status, and that TET1 methylation might be an earlier event than hMLH1 methylation in the precancerous condition. In addition, TET1 may be a frequent target of CIMP-related hypermethylation rather than the cause of CIMP and contribute to the establishment of pathway displaying extensive DNA methylation in tumorigenesis.

Regarding links to clinicopathological variables, colorectal cancers with TET1 methylation, but not hMLH1 methylation, showed significantly less metastatic events. Since increased frequency of DNA methylation in a certain set of genes can determine the behavior of these tumors, it may be possible that colorectal cancers with TET1 methylation may exhibit unique clinicopathological features compared to those without TET1 methylation.

In conclusion, we demonstrate here, for the first time, the biological and clinical significance of TET1 in colorectal cancers associated with CIMP. The understanding of these multiple pathways towards colorectal cancer has implications for therapy. Recent combined genetic and epigenetic analyses suggest that colorectal cancer is not a uniform disease (1, 24, 34). The distinct clinicopathological features in relation to CIMP raise important implications for prevention, screening, and therapy. We provide new findings here showing a link between TET1 methylation and enhanced DNA methylation in a subset of tumors, which may be associated with unique clinicopathological features of this disease. Now, individualized therapy for colorectal cancer is required, whereby each tumor may be examined via molecular profiling before therapy can be selected. Our findings provide evidence that TET1 methylation may be a good biomarker for the prediction of metastasis in colorectal cancer.

No potential conflicts of interest were disclosed.

Conception and design: N. Ichimura, K. Katsushima, Y. Kondo

Development of methodology: N. Ichimura, K. Katsushima, Y. Kondo

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): N. Ichimura, K. Shinjo, B. An, Y. Shimizu, M. Tojo, E. Yamamoto, H. Suzuki

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): N. Ichimura, K. Shinjo, A. Hatanaka, M. Tojo, Y. Kondo

Writing, review, and/or revision of the manuscript: N. Ichimura, Y. Kondo

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): N. Ichimura, F. Ohka, Y. Kondo

Study supervision: N. Ichimura, K. Yamao, M. Tojo, E. Yamamoto, H. Suzuki, M. Ueda, Y. Kondo

This work is supported by Grant-in-Aid for Scientific Research from the Japan Society for the Promotion of Science (25290048, Y. Kondo), PRESTO of JST (Y. Kondo), P-Direct of Ministry of Education, Culture, Sports, Science and Technology (Y. Kondo), and Takeda Science Foundation (Y. Kondo).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.