Myeloid Differentiation Factor 88 Signaling in Bone Marrow–Derived Cells Promotes Gastric Tumorigenesis by Generation of Inflammatory Microenvironment

Gastric cancer is the second leading cause of death from malignancy worldwide (1). Helicobacter pylori (H. pylori) infection induces chronic gastritis, which is closely associated with gastric cancer (2, 3). It has been shown that COX-2 expression is induced by H. pylori infection in the stomach (4), and that the activation of downstream signaling by prostaglandin E₂ (PGE₂) plays a key role in tumorigenesis in a variety of organs (5). We previously showed that induction of COX-2/PGE₂ pathway in the gastric mucosa of K19-C2mE mice by expression of Ptgs2 and Ptges causes inflammatory responses in the glandular stomach (6). Moreover, Wnt signaling activation in the gastric mucosa of K19-Wnt1 mice by expression of Wnt1 results in the development of limited dysplastic lesions; however, the simultaneous activation of Wnt signaling and COX-2/PGE₂ pathway in Gan mice leads to development of inflammation-associated gastric tumors (7, 8). These results indicate the role of COX-2/PGE₂ pathway-induced inflammation in gastric tumorigenesis. In the gastric tumor stroma, proinflammatory cytokines including TNFα, IL1β, and IL11 are expressed and promote tumorigenesis through enhancement of undifferentiated status, recruiting of myeloid-derived suppressor cells and increasing tumor cell survival (9–11).

On the other hand, it has been shown that innate immune responses through Toll-like receptors (TLRs) promote tumor development through the regulation of inflammatory and immune responses in tumor tissues (12, 13). Mouse genetics studies have indicated that signaling through myeloid differentiation factor 88 (MyD88), which is an effector molecule of the TLRs, is required for the development of sporadic intestinal tumors (14–16), colitis-associated colon cancer (17, 18), gastric cancer (19), liver cancer (20), skin tumors, and sarcomas (21). MyD88 has also been shown to promote tumor metastasis (22). Moreover, MyD88 signaling induces the expression of COX-2 and inflammatory cytokines in tumor tissues, which contribute to the inflammatory microenvironment generation (14, 17, 18, 20). However, relationship between the COX-2/PGE₂ pathway and TLR/MyD88 signaling in the induction of inflammatory responses and the promotion of tumorigenesis is not yet fully understood.

Here, we examined the role of MyD88 in gastric tumorigenesis using Gan mice, in which COX-2/PGE₂ pathway is constitutively activated. Importantly, we show that MyD88 signaling in BMDCs is required for the generation of the inflammatory microenvironment in tumors even though the COX-2/PGE₂ pathway is activated. Moreover, inflammatory responses induce the expression of TLR2 and CD14 in tumor epithelial cells, which have been shown to play tumor-promoting roles through enhancement of stem cell properties (16, 23, 24). These results indicate that both the MyD88 signaling and the COX-2/PGE₂ pathway are required for generation of the inflammatory microenvironment, which may promote tumorigenesis through the induction of TLR2/CD14 signaling in tumor cells. These results suggest that targeting both TLR/MyD88 signaling and the COX-2/PGE₂ pathway will be an effective preventive strategy for gastric cancer.

The construction of Gan mice, the gastric tumor model that was used in the current study, is described in Supplementary Information. Myd88-mutant mice were purchased (Oriental BioService), and Irf5-mutant mice were constructed as described previously (25). For the gastric tumor phenotype analyses, each genotype mice were euthanized and examined at 40 to 50 weeks of age. For the preneoplastic lesion analyses, K19-Wnt1 mice were examined at 45 to 50 weeks of age. To induce gastric mucosal regeneration, tamoxifen (150 mg/kg) was administered to 10-week-old wild-type mice by oral gavage for 5 continuous days. This resulted in a loss of parietal cells, leading to mucosal regeneration (26). All of the mice used in this study were born and raised in the same room of the Kanazawa University SPF facility (Kanazawa, Japan). The protocols of all animal experiments were approved by the Committee on Animal Experimentation of Kanazawa University (Kanazawa, Japan).

Gan mouse tumors are not polypotic, thus the numbers cannot be counted. Thus, the tumor height was measured using histologic sections, and the relative tumor height was compared with the control. The number of preneoplastic lesions of K19-Wnt1 mouse stomach was counted using six independent histologic sections from each mouse, and the mean number of lesions per section was calculated.

Tissues were fixed in 4% paraformaldehyde, paraffin-embedded, and sectioned at 4-μm thickness. The sections were stained with haematoxylin and eosin (H&E) or processed for IHC. The antibodies and visualization methods are described in Supplementary Information. The mean index for Ki-67 or F4/80 was calculated by counting the labeled cells per microscopic field at 200× in five fields. The immunostaining-positive area in the field was measured using ImageJ 1.48 software program (NIH, Bethesda, MD). Apoptotic cells were detected using an ApopTag Peroxidase In Situ Apoptosis Detection Kit (Millipore).

Total RNA was extracted from the mouse gastric tumors and the normal wild-type mouse stomach (n = 3 for each genotype) using ISOGEN (Nippon Gene), and purified using an RNeasy Mini Kit (Qiagen). Inflammation-related genes were analyzed using the RT² Profiler PCR Array Mouse Innate & Adaptive Immune Responses (Qiagen). The data were analyzed using the RT² Profiler PCR Array Data Analysis version 3.5 software (Qiagen) and were compared with the wild-type mice.

Tissues were homogenized in lysis buffer and the supernatant protein sample was separated in SDS-polyacrylamide gel. The antibodies are described in Supplementary Information. The ECL Detection System (GE Healthcare) was used to detect the signals.

Bone marrow cells were prepared from the femurs and tibias of donor mice. Recipient mice were irradiated (9 Gy), followed by an intravenous injection of 2 × 10⁶ bone marrow cells. The x-ray CT images of the gastric tumors were examined using a LaTheta LCT-100 instrument (Aloka) at every 4 weeks until 20 weeks after bone marrow transplantation. The tumor areas of the slice images were measured using the ImageJ 1.48 software program (NIH) as described previously (9).

Total RNAs were extracted from cells or tumor tissues using ISOGEN (Nippon Gene), reverse-transcribed using the PrimeScript RT reagent kit (Takara), and PCR-amplified by a Stratagene Mx3000P instrument (Agilent Technologies) using SYBR Premix ExTaqII (Takara). The primers used for the real-time RT-PCR except for Cd14 were purchased from Takara, and the primer IDs and sequences for Cd14 are shown in Supplementary Table S1.

The human gastric cancer cell lines, AGS (ATCC), MKN74 (RIKEN BioResource Center, Japan), SNU484 [Korean Cell Line Bank (KCLB)], and the mouse macrophage cell line, RAW264 (RIKEN BioResource Center, Japan), were cultured in RPMI1640 medium supplemented with 10% FBS. Cell lines were authenticated by isoenzyme analysis or short tandem repeat by the ATCC, RIKEN, and KCLB. All cell lines were initially expanded and cryopreserved within 1 month of receipt. Cells were used for 3 months after thawing frozen vials. Construction of knockdown cell lines and the preparation of the conditioned medium (CM) are described in Supplementary Information.

Total RNA was extracted from the mouse gastric tumors and the normal wild-type mouse stomach (n = 3 for each genotype) using ISOGEN (Nippon Gene) and purified using an RNeasy Mini Kit (Qiagen). A sequencing analysis and upstream pathway analysis were performed according to the methods described in Supplementary Information. The sequence results were deposited in the Gene Expression Omnibus, as accession GSE70135.

The data were analyzed using the unpaired t test and are presented as the mean ± SE or SD. A value of P < 0.05 was considered to be statistically significant.

Innate immune signaling through TLR/MyD88 has been shown to regulate immune responses in tumors (12, 13). IFN regulatory factor 5 (IRF5) has also been shown to play a role in the TLR/MyD88 pathway through the induction of proinflammatory cytokines, including TNFα (25). Because TNFα plays a tumor-promoting role in Gan mice (9), we examined the roles of both MyD88 and IRF5 in the current study. The expression of both MYD88 and IRF5 was significantly increased in human intestinal-type gastric cancer, and significant upregulation of Myd88 was also found in Gan mouse gastric tumors (Supplementary Fig. S1). We thus crossed Gan mice with Myd88 and Irf5 gene knockout mice to generate Gan Myd88^−/− and Gan Irf5^−/− mice, in which Myd88 or Irf5 was disrupted in both epithelial and stromal cells, respectively. Although the control Gan mice developed large gastric tumors by 40 to 50 weeks of age, gastric tumorigenesis was significantly suppressed in the Gan Myd88^−/− mice, but not in the Gan Irf5^−/− mice (Fig. 1A). The mean tumor sizes in the Gan Myd88^−/− mice were 49% of the control Gan mice (Fig. 1B). These results indicate that MyD88, but not IRF5, plays a role in gastric tumor development.

Apoptotic cells were found on the tumor surface in both the control Gan and the Gan Irf5^−/− mice, and similar apoptotic cell localization was found in the Gan Myd88^−/− mice, indicating that MyD88 pathway does not affect apoptosis (Fig. 1C). In contrast, Ki-67–labeled cells were found throughout the tumor tissues of Gan and Gan Irf5^−/− mice, whereas they were mostly limited to the proliferating zone at the gland neck in the Gan Myd88^−/− mice. Consistently, the Ki-67–labeling index was reduced, to a significant extent, in the Gan Myd88^−/− mouse tumors, suggesting that MyD88 signaling promotes tumor cell proliferation (Fig. 1D).

In Gan mouse gastric tumors, macrophages infiltrated the tumor stroma and generated an inflammatory microenvironment (Fig. 2A). In the Gan Myd88^−/− mouse tumors, the number of macrophages was significantly decreased in comparison to Gan and Gan Irf5^−/− mouse tumors (Fig. 2A and B).

We thus performed PCR array analyses to examine the expression of inflammation-related genes in the respective genotype mouse tumors. In the control Gan mouse tumors, expression levels of proinflammatory cytokines Il17a, Tnf, Il1a, Il1b, Il6, and Il23a, chemokine receptors Ccr4 and Ccr6, and TLRs, Tlr1, Tlr2, and Tlr5 were significantly increased (Fig. 2C and D). Although the upregulation of Il1b, Il6, Ccr4, and Ccr6 was suppressed in Gan Irf5^−/− mouse tumors, most other inflammation-related genes were upregulated, suggesting that IRF5 signaling may be less involved in the inflammatory responses in gastric tumors. Importantly, in the Gan Myd88^−/− mouse tumors, induction of inflammation-related genes was significantly suppressed, indicating that MyD88 signaling is essential for inflammatory microenvironment generation (Fig. 2C and D).

We also found substantially decreased levels of phosphorylated IκB and NF-κB in Gan Myd88^−/− mouse tumors (Fig. 2E). Moreover, we confirmed by IHC that NF-κB is activated in macrophages of Gan mouse tumors, which was suppressed in Gan Myd88^−/− mice (Fig. 2F). In Gan mouse tumors, the COX-2/PGE₂ pathway is constitutively activated by transgenic expression of Ptgs2 and Ptges encoding COX-2 and mPGES-1, respectively; we confirmed the expression of COX-2 and mPGES-1 in Gan Myd88^−/− mouse tumors. In two Gan Myd88^−/− mice, mPGES-1 levels were decreased in comparison to control, possibly due to the decrease in endogenous mPGES-1 levels by suppression of inflammation (Fig. 2G). We previously showed that Il1r1 disruption did not affect the inflammation phenotype in K19-C2mE mice, indicating that the IL1R/MyD88 pathway is not required for COX-2/PGE₂ pathway-induced inflammation (27). Accordingly, these results indicate that TLR/MyD88 signaling is required for the generation of the COX-2/PGE₂ pathway-induced inflammatory microenvironment through the activation of NF-κB. Interestingly, the level of active β-catenin was also decreased by Myd88 disruption (Fig. 2E), suggesting that MyD88 has a role in activation of Wnt signaling in tumor cells (see below).

We next examined the expression of MyD88-dependent genes in another inflammation-associated gastric tumor model, gp130^F/F mice, in which hyperactivation of Stat3 promotes gastric tumorigenesis via the TLR2/MyD88 cascade (19, 28, 29). The expression levels of Tnf, Cd14, Casp1, Il1a, and Tlr2 were significantly increased in gp130^F/F tumors; however, this upregulation was not suppressed in the gp130^F/F Myd88^−/− mouse tumors, in which Myd88 was disrupted in both epithelial and stromal cells (Supplementary Fig. S2). It is therefore possible that the hyperactivation of Stat3 (in gp130^F/F mice) can mimic the role of COX-2/PGE₂ and MyD88 (in Gan mice) in the generation of the inflammatory microenvironment; however, this possibility remains to be investigated.

To examine the role of MyD88 in BMDCs in gastric tumorigenesis, we next performed bone marrow transplantation from Myd88^−/−, Irf5^−/−, or GFP transgenic mice into Gan mice to generate the respective bone marrow chimeric mice. In these models, Myd88 or Irf5 was disrupted in the BMDCs but not in the epithelial cells of the tumor tissues. Chronologic X-ray CT analyses showed a significant increase in the gastric tumor volume of the bone marrow–transplanted (BMT)-Gan mice that received bone marrow from GFP or Irf5^−/− mice during the 20-week observation period (Fig. 3A and B). In contrast, gastric tumor growth was significantly suppressed in the BMT-Gan mice that received Myd88^−/− mouse bone marrow. Macrophage infiltration into the tumor tissues was significantly decreased by the disruption of Myd88 in BMDCs (Fig. 3C and D). We further examined the gene expression in gastric tumors of BMT-Gan mice by next-generation sequencing. We found that the levels of inflammatory cytokines and chemokines were decreased in the BMT-Gan from Myd88^−/− mice in comparison with the control Gan mice (Fig. 3E). Consistently, NF-κB activation was suppressed in macrophages by the disruption of Myd88 in BMDCs (Fig. 2F). Taken together, these results indicate that MyD88 signaling in BMDCs is important for the inflammatory microenvironment generation, which may play a tumor-promoting role.

Among the MyD88-dependent upregulated genes in Gan mouse tumors, we focused on TLR2 and its coreceptor CD14 because we had previously demonstrated that epithelial TLR2 expression is important for gastric tumorigenesis in gp130^F/F mice (28). We found that the expression levels TLR2 and CD14 were increased significantly in human gastric cancer and Gan mouse tumor tissues which were a mixture of epithelial and stromal cells (Supplementary Fig. S3). We thus performed laser microdissection–based RT-PCR, and found that the expression of Tlr2 and Cd14 in Gan mouse tumors was predominantly detected in E-cadherin (Cdh1)-expressing tumor epithelial cells rather than stromal cells (Fig. 4A). These results suggest that the inflammatory microenvironment induces the expression of Tlr2 and Cd14 in tumor epithelial cells.

To further confirm the inflammation-dependent induction of TLR2 and CD14 in tumor cells, we treated gastric cancer cells with CM from activated RAW264 macrophages treated with LPS. Importantly, we found that the expression levels of TLR2 and CD14 were significantly increased in AGS, MKN74, and SNU484 cells after treatment with LPS-stimulated RAW264 CM but not with control CM (Fig. 4C). Moreover, the induction of TLR2 and CD14 in gastric cancer cells was significantly suppressed when Myd88 expression was knocked down by CRISPR/Cas9 system (KD-Myd88) in RAW264 cells (Fig. 4B and C). These results indicate that MyD88-dependent macrophage-derived factors upregulate TLR2 and CD14 in cancer cells. Moreover, the expression of stem cell marker CD44 was significantly increased in AGS cells when the cells were treated with LPS CM, whereas it was suppressed by knockdown of Myd88 expression in RAW264 macrophages (Fig. 4D). These results indicate that MyD88-dependent macrophage-derived factor(s) induces the TLR2/CD14 signaling pathway in cancer cells, which may promote tumorigenesis by enhancing the undifferentiated status of the tumor cells.

Tamoxifen treatment induces loss of parietal cells in mouse stomach, which leads to expansion of undifferentiated cell population and regeneration of the gastric mucosa (Supplementary Fig. S4) as described previously (26). Notably, the expression levels of Tlr2 and Cd14 were significantly increased in these regenerating gastric mucosa, suggesting a role for TLR2/CD14 signaling in undifferentiated epithelial cells in the stomach.

We next examined the differentiation status of the gastric tumors of Gan Myd88^−/− mice, in which Myd88 was disrupted in both epithelial and stromal cells. Importantly, the expression levels of intestinal stem cell markers, Cd44, Ephb3, Cd133, and Sox9 (30), were significantly decreased in the Gan Myd88^−/− mouse tumors but not in Gan Irf5^−/− mice compared with the control Gan mice (Fig. 5A). CD44 expression was not detected in the normal glandular stomach, however, CD44-expressing cells were found throughout the tumor tissues of Gan mice (Fig. 5B). In contrast, there were substantial decreases in the numbers of CD44-positive cells in Gan Myd88^−/− mice and BMT-Gan from Myd88^−/− mice (Fig. 5B). We also found that the expression of MUC5AC and H⁺K⁺/ATPase (differentiation markers) was not expressed in control Gan mouse tumors, but induced in the tumors of Gan Myd88^−/− and BMT-Gan from Myd88^−/− mice (Fig. 5B). Notably, there are chimeric glands consisting of both CD44-positive cells and H⁺K⁺/ATPase-positive cells in BMT-Gan from Myd88^−/− mice, suggesting that tumor epithelial cells are differentiated by Myd88 disruption in BMDCs (Fig. 5B, white arrowheads).

We performed next-generation sequencing using the tumor tissues of Gan Myd88^−/− mice and BMT-Gan from Myd88^−/− mice to examine the gene expression profiles of the respective mouse tumors. In the Gan Myd88^−/− mouse tumors, Myd88 was disrupted in both the epithelial and stromal cells, whereas Myd88 was disrupted only in the BMDCs in the BMT-Gan from Myd88^−/− mice. We found that the numbers of genes that were downregulated below 0.5-fold or upregulated above 2.0-fold in tumors in both Gan Myd88^−/− mice and BMT-Gan from Myd88^−/− mice compared with control Gan mouse levels were 908 and 318, respectively (Fig. 6A). Using the sequencing results, we performed an Ingenuity Pathway Analysis (IPA) to examine the pathways that were inhibited by disruption of Myd88 in BMDCs. We found that NF-κB pathway was the most inhibited by Myd88 disruption, and inflammatory cytokine pathways were also significantly inhibited (Fig. 6B and Supplementary Fig. S5). Notably, the IPA analysis showed that the Wnt/β-catenin (CTNNB1) signaling was also significantly suppressed by Myd88 disruption, which was consistent with the immunoblotting results (Fig. 2E). These results suggest that MyD88 signaling in BMDCs promotes tumorigenesis through activation of Wnt signaling in tumor cells.

We previously showed that macrophages were infiltrated in the stroma of preneoplastic lesions of the K19-Wnt1 mouse stomach, and that Wnt signaling activity was promoted in epithelial cells of these lesions, suggesting that macrophage-derived factors activate Wnt signaling (31). Notably, the numbers of preneoplastic lesions in K19-Wnt1 Myd88^−/− mice were significantly decreased in comparison with control K19-Wnt1 mice (Fig. 6C and D). It is therefore possible that the MyD88 signaling in BMDCs contributes to promotion of Wnt signaling activity in the tumor epithelial cells, which may contribute to the acceleration of the undifferentiated status and tumorigenesis.

Taken together, the current study strongly suggests that MyD88 signaling in BMDCs together with COX-2/PGE₂ pathway promote tumorigenesis through the induction of the TLR2/CD14 pathway as well as activation of Wnt signaling in tumor cells.

It has been established that chronic inflammation plays an important role in cancer development through a variety of mechanisms (32, 33), and that the COX-2/PGE₂ pathway plays a key role in the promotion of tumorigenesis through the generation of the inflammatory microenvironment (5, 34–36). Notably, the induction of COX-2 expression in intestinal tumors was suppressed in Apc^Min Myd88^−/− mice (14), suggesting that MyD88 signaling induces COX-2 expression in tumor tissues possibly through NF-κB activation, which induces inflammatory responses. However, in the current study, we have shown that the inflammatory responses are significantly suppressed by Myd88 disruption in BMDCs, even though the COX-2/PGE₂ pathway was constitutively activated. It is therefore possible that the simultaneous activation of both MyD88 signaling in BMDCs and the COX-2/PGE₂ pathway is required to generate the inflammatory microenvironment in tumor tissues (Fig. 6E). We previously showed that IL1R signaling is not required for the maintenance of COX-2/PGE₂ pathway-induced inflammatory responses (27), thus TLR/MyD88 rather than IL1R/MyD88 pathway plays a role in the generation of inflammatory microenvironment in gastric tumors.

It has consistently been shown that the disruption of Myd88 or Tlr2, 4, 9 in the BMDCs of Apc-mutant mice caused a reduction in cytokine expression and the suppression of intestinal tumorigenesis (37). Moreover, it has been demonstrated that the IKKβ-induced NF-κB activation in myeloid cells is important for inflammatory responses and proliferation in the stomach after Helicobacter infection (38). Taken together, these results indicate that the TLR/MyD88/NF-κB signaling in BMDCs plays a key role in generating the inflammatory microenvironment in tumor tissues.

It is intriguing to determine the specific TLRs and ligands that are required for the activation of the MyD88/NF-κB pathway in BMDCs in tumor tissues. TLR2 and TLR4 have been demonstrated to be important for the homeostasis of intestinal epithelial cells (39), and polymorphisms of TLR2 and TLR4 have been shown to be associated with an increased risk of gastric cancer (40, 41). Accordingly, TLR2 or TLR4 are important candidates for responsible TLRs in gastric tumorigenesis. TLRs recognize not only pathogen-associated molecular patterns (PAMP) but also endogenous ligands such as damage-associated molecular patterns (DAMP; refs. 13, 42). It has been reported that TLR2/4 recognize high-mobility group box 1 and extracellular matrix proteoglycan versican that are produced by cancer cells (24, 43, 44). Accordingly, cancer cell–derived ligands as well as infectious agents may activate TLR2 and/or TLR4 in tumor tissues, thereby generating the inflammatory microenvironment.

We also showed that inflammatory responses induced the expression of TLR2 and CD14 in tumor epithelial cells (Fig. 6E). We previously showed that TLR2/MyD88 signaling in tumor cells promotes gastric tumorigenesis in gp130^F/F mice (19, 28). Consistent with these findings, it has been shown that TLR2 signaling in cancer cells plays a tumor-promoting role through the enhancement of stemness (23, 24). Moreover, the epithelial cell–specific deletion of Myd88 was observed to suppress intestinal tumorigenesis in Apc^Min mice (16). These results suggest that the TLR2/CD14 signaling in “epithelial cells” promotes tumor development through the maintenance of the undifferentiated status of tumor cells. Notably, it has been reported that H. pylori infection activates TLR2 but not TLR4 (45). Accordingly, it is possible that H. pylori infection can promote gastric tumorigenesis through the activation of TLR2 on tumor epithelial cells. On the other hand, the NF-κB pathway is activated in both differentiated (tubular type and papillary type) and undifferentiated (mucinous type) human gastric cancers, which suggests that the MyD88/NF-κB pathway is less involved in the regulation of the histologic subtypes of established gastric cancer (Supplementary Table S2).

The expression of Tlr2 and Cd14 in intestinal and mammary epithelial cells is associated with the activation of Wnt signaling, which may explain the increased stemness and tumorigenicity (16). We herein showed that Wnt/β-catenin signaling is significantly suppressed and that differentiation is induced in gastric tumors by Myd88 disruption, supporting the idea that TLR2/CD14 signaling enhances the stemness of tumor cells through activation of Wnt signaling. It has been reported that NF-κB activation promotes Wnt signaling in the intestinal epithelial cells, which causes dedifferentiation and the acquisition of stem cell properties (46). It is therefore possible that the activation of NF-κB by TLR2/CD14/MyD88 signaling increases the Wnt signaling activity in tumor epithelial cells, thereby contributing to the maintenance of undifferentiated status of tumor cells (Fig. 6E).

In the current study, we showed that TLR/MyD88 signaling in BMDCs is required for the generation of the inflammatory microenvironment in cooperation with the activation of the COX-2/PGE₂ pathway. MyD88-dependent macrophage-derived factors in the inflammatory microenvironment induce the expression of TLR2 and CD14 in tumor cells. Wnt signaling activation is a possible mechanism underlying the TLR2/CD14/MyD88 signaling–induced tumor promotion. On the basis of these findings, we propose that the targeting of TLR/MyD88 signaling together with the COX-2/PGE₂ pathway will be an effective preventive strategy against gastric cancer.

H. Saya reports receiving commercial research grants from Daiichi Sankyo Co., Ltd. and Eisai Co., Ltd. No potential conflicts of interest were disclosed.

Conception and design: M. Oshima

Development of methodology: Y. Maeda

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Y. Maeda, K. Echizen, H. Oshima, L. Yu, N. Sakulsak, B.J. Jenkins

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Y. Maeda, K. Echizen, H. Oshima, O. Hirose, Y. Yamada, T. Taniguchi, B.J. Jenkins

Writing, review, and/or revision of the manuscript: B.J. Jenkins, M. Oshima

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): T. Taniguchi, H. Saya

Study supervision: H. Saya, M. Oshima

The authors thank Manami Watanabe and Ayako Tsuda for technical assistance. The computations were partially performed on the NIG supercomputer at the ROIS National Institute of Genetics, Japan.

This work was supported by AMED-CREST, AMED, Japan Agency for Medical Research and Development, Japan (to M. Oshima), and Grants-in-Aid for Scientific Research on Innovative Areas from the Ministry of Education, Culture, Sports, Science and Technology of Japan (#22114005 and #15H02362; to M. Oshima). This work was also supported in part by the National Health and Medical Research Council of Australia (to B. Jenkins) as well as the Operational Infrastructure Support Program by the Victorian Government of Australia (to B. Jenkins).

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.