Systemic DC Activation Modulates the Tumor Microenvironment and Shapes the Long-Lived Tumor-Specific Memory Mediated by CD8⁺ T Cells

Cancer immunotherapy has recently enjoyed a renaissance as a result of new and more effective approaches (1). For example, adoptive T-cell transfer with tumor antigen–specific TCR or chimeric antigen receptor (CAR) gene-transferred T cells has shown antitumor effects on some cancers (2, 3) Also, blocking immune checkpoints using mAbs to CTL–associated protein-4 (CTLA-4) or programmed cell death protein-1 (PD-1) has demonstrated clinical efficacy (4). Another promising approach involves active immunization using a cancer vaccine, and in this context the function of dendritic cells (DC) appears to be very important (5, 6). The next breakthroughs for its use as an effective antitumor immunotherapy would be the induction of memory T cells and the modulation of tumor microenvironment in the light of systemic and local immune responses. Once established, memory T cells could quickly respond to emerging cancer cells in the recurrence and metastasis. The persistence of functional memory T cells was desired for immunotherapies. In addition, therapeutic options to overcome the barriers of tumor microenvironment and to support T-cell infiltration and function in tumors are required. For this perspective, organized lymphoid aggregates found locally at tumor sites, termed tertiary lymphoid structures (TLS), should be a major focus that may lead effector T cells to recruit to tumor cells (7, 8).

For the success of immunotherapy, it has been recently argued that combining multiple therapies should be beneficial, because different types of immune responses may prevent tumor cells from immune escape. In this point, a strategy linking innate and adaptive immune responses is attractive and promising, for example, they can attack MHC⁺ or MHC⁻ tumor cells. NKT cells have bipotential capacity and can suppress or activate immune responses. Once activated by their ligands, they can closely contact DCs and mature them systemically in lung, liver, spleen, and bone marrow. DCs thus licensed by activated NKT cells via both CD40L on NKT cells and their simultaneous production of inflammatory cytokines, for example, IFNγ and TNFα (9). Then, the DCs can induce adaptive immunity in infection and tumor models (10–13). We and others have shown that coadministration of NKT cell ligand and antigen can induce adaptive immunity (10, 14, 15).

On the basis of these evidences, we previously illustrated the concept of a unique approach of using NKT ligand–loaded CD1d⁺ allogeneic cells transfected with tumor antigen mRNA as artificial adjuvant vector cells (aAVC; refs. 16, 17). The first generation of aAVCs demonstrated certain of its immunologic features inducing both innate and adaptive immunity in the lymphoid organs through antigen-captured DCs in situ. To develop the aAVC therapy, we assumed that the optimal amount of target protein delivered to in vivo DC would affect the magnitude of the ensuing immune responses in the tumor. By the new method, aAVCs can produce a suitable amount of protein for protein-based clinical immunotherapy and be developed as a current type of aAVC. Here we assessed the antitumor effects and further analyzed the cellular mechanism against tumor in terms of local and systemic immune responses. This study not only demonstrates the efficacy of our approach, but will also help to identify the key components needed for successful future therapeutics.

Human IL2 was purchased from Shionogi & Co., LTD. Human recombinant IL7 and IL15 were purchased from Peprotech, Inc. The clinical grade of α-GalCer was provided from Regimmune. OVA_257–264 peptide (SIINFEKL) and WT1_235–243 (CMTWNQMNL) were obtained from Toray Research Center, Inc. The antibodies used in this study were purchased and described in Supplementary Table S1. OVA-tetramer was purchased from MBL. WT1-tetramer (HLA-A24) and HIV-tetramer (HLA-A24) were produced by our group. A FACSCalibur or FACSCanto II instrument and CELLQuest, Diva (BD Biosciences), and FlowJo (Tree Star) software were used for the analysis. Anti-PD-1 mAb (RPMI-14) was produced as described previously (18). Recombinant human WT1 was synthesized in RIKEN.

B16 and HEK293 cell lines were purchased from ATCC. NIH3T3 and WEHI3B cell lines were obtained from the RIKEN Cell Bank. MO4 (19) and J558 (20) cell lines were received from Dr. R.M. Steinman (The Rockefeller University, New York, NY) and has been routinely tested for OVA expression or for IgA and H2-Kd expression by our hands, respectively. The J558-WT1 cell line was established by transfection of a human WT1 cDNA expression vector into J558. All cell lines were tested according to the manufacturer's protocol and proved to be mycoplasma free (Mycoplasma Detection Kit; Minerva Biolabs). C57BL/6 or BALB/c mice were purchased from CLEA Japan. Ly5.1 congenic OT-1 mice were generated by cross/backcross breeding of OT-1 with B6. Ly5.1 mice and screening for the presence of Vα2 and Ly5.1 and absence of Ly5.2 by flow cytometry. OT-1 TCR transgenic mice, CD11c-DTR/GFP mice (21), and XCR1-DTR-venus mice (22) and other all the mice were maintained under specific pathogen-free conditions and studied in compliance with our institutional guidelines.

Human PBMCs were isolated from healthy volunteers. All studies were approved by the RIKEN institutional review board. Murine bone marrow–derived DCs were generated in the presence of GM-CSF for 6 days as described previously (23). The preparation of aAVCs using HEK293 or NIH3T3 was performed as described previously (17). Briefly, cells resuspended in OptiMEM and RNA were transferred to a cuvette and then the cell suspension was pulsed in ECM 830 Square Wave Electroporation System (Harvard Apparatus). Pulse condition was a single 500 V, 3 ms square pulse. Immediately after electroporation, the cells were transferred to culture medium and cultured in the presence of 500 ng/mL of α-GalCer. The protein expression of transfected cells was analyzed by ELISA (ITEA) for OVA, flow cytometry for CD1d, and Western blot analysis for WT1 or TRP-2 protein as described previously (17).

Antigen-specific IFNγ-secreting cells were performed by ELISPOT assay as described previously (24). The cytotoxic activity of NKT cell line was analyzed using LDH assay kit according to the manufacturer's instructions (Takara Bio Company). Expression of immune response–related molecules in DCs and tumors was analyzed by quantitative PCR assay. In the analysis of splenic DCs and tumor, the total RNA was isolated using RNeasy Mini Kit (Qiagen) and cDNA was synthesized by ReverTra Ace (Toyobo) according to the manufacturer's instructions. Tumor-associated DCs were sorted and directly subjected to cDNA synthesis using a CellsDirect One-Step qRT-PCR Kit (Invitrogen) with a mixture of pooled gene-specific primers (Supplementary Table S2). Synthesized cDNA was appropriately diluted with water and used as template for subsequent quantitative PCR as described previously (25).

The P values were calculated with the two-side Student t test or the Mann–Whitney U test. The log-rank test was used for survival calculations. P < 0.05 was considered statistically significant.

We previously established aAVCs, which are α-GalCer-loaded and antigen mRNA-transfected, CD1d-expressing allogeneic cells (NIH3T3 or HEK293 cells for mice and humans, respectively; ref. 17). In the current study, using a developed method by optimizing several points of the protocol to increase protein production (Supplementary Fig. S1A), we succeeded in the production of target antigen protein by aAVC at levels about 100 times higher than with the previous method. As shown in Supplementary Fig. S1B–S1D, we verified that aAVC-OVA cells, which were loaded with α-GalCer and cotransfected with OVA and CD1d mRNA, highly expressed the CD1d molecule and produced abundant OVA protein, and could directly stimulate NKT cells, but not OT-1 T cells in vitro (17). In the initial study, we assessed antitumor effects in a therapeutic model. We administered MO4 cells (OVA-expressing B16 melanoma) and, after verifying that large tumors (tumor size: 441.7 ± 176.8 mm³) were established, mice were treated with aAVC-OVA. As shown in Fig. 1A, untreated tumors grew larger; however, they stopped growing after the aAVC-OVA treatment and the center of tumor mass became necrotic. Given this striking immunotherapeutic outcome, we began to analyze the antitumor effects in more detail. This initial study thus demonstrated that treatment with aAVC-OVA increased the frequency of tumor-infiltrating T cell, especially CD8⁺ T cells in spleen and the tumor (Fig. 1B and Supplementary Fig. S2).

We and others previously demonstrated that in vivo DCs stimulated by activated NKT cells produce IL12 and certain chemokines for inducing adaptive immunity (9, 10, 15, 24). In the current study, we analyzed the DCs in vivo in more detail after administration of aAVC. Both CD8a⁺ and CD8a⁻ subsets of DCs in the spleen highly expressed CD40, CD80, and CD86 16 hours after aAVC treatment, whereas 40 hours later, CD8a⁺ DCs also highly expressed crucial molecules for the generation of memory T-cell response, such as CD70, IL15Ra, and 4-1BBL (Supplementary Fig. S3A). In addition, phosphorylation of both STAT1, the signal transducer for type I and II IFNs, and STAT5, the signal transducer for IL2, IL7, IL15, and GM-CSF, was observed in both subsets of DCs 5 hours later (Fig. 1C). DCs could also be seen to have taken up debris of aAVC and to undergo maturation after capturing aAVC (Supplementary Fig. S3B). XCR1⁺ DCs is the relevant of CD8a⁺ subset of CD11c⁺ splenic DCs and are responsible for cross-presentation of exogenous antigen (22). We confirmed that XCR1 is exclusively expressed among CD8a⁺CD11c⁺MHCII⁺ DCs in the spleen and CD8a⁺ CD11b⁻CD11c⁺MHCII⁺ DCs (resident DCs) and CD103⁺CD11b⁻CD11c⁺MHCII⁺ DCs (migratory DCs) in lymph node (Supplementary Fig. S4A and S4C). In fact, XCR1⁺ DCs are specifically depleted by diphtheria toxin (DT) treatment (Supplementary Fig. S4B and S4D). To test the cross-presenting activity of the DCs in aAVC-immunized mice, we used WT, CD11c-DTR, and XCR1-DTR mice. These mice were transferred with CFSE-labeled OT-1 T cells and then injected with aAVC-OVA on the following day. Three days later, the OT-1 cells showed significant proliferation in WT mice, but not in CD11c-DTR or XCR-DTR mice treated with DT (Fig. 1D). These findings indicated that the CD8a⁺ DCs that had engulfed aAVC-OVA were activated and able to cross-present OVA peptide to CTLs. Next, we sorted the CD8a⁺ DCs and evaluated their expression of chemokines, cytokines, and genes involved in antigen processing, molecules that would play a key role in antigen presentation. We detected significantly elevated expression of CCL17, CCL22, IL12a, IL15, IL27, TAP1, TAP2, and PA28, a component of the immunoproteasome (Fig. 1E). Taken together, aAVC therapy can deliver tumor antigens to in vivo DC.

Because the aAVC therapy induced tumor regression, we next focused on the tumor microenvironment. The expression of CXCL9, CXCL10, and CXCL11 were upregulated in tumors of aAVC-treated, but not untreated mice, as assessed by real-time PCR (Fig. 2A). Interestingly, CXCL10 was apparently expressed on DCs in tumor site (Fig. 2B). We verified it at protein level by IHC (Fig. 2C). Next, to focus on the expression of CXCR3 that are receptors for CXCL10, we detected downregulation of CXCR3 on OVA-tetramer⁺CD8⁺ T cells in tumor but not in spleen (Fig. 2D), suggesting that the activation of this receptor. We further analyzed CD8⁺ T cells and DCs and observed that many CD8⁺ T cells were located close to the CD31⁺ tumor vessels along with CD11c⁺ DCs in the aAVC-injected mice but not in untreated mice (Figs. 1B and 2E). In addition, the blood vessels in the tumors of the treated mice were CD31⁺VCAM-1⁺ICAM-1⁺ (Fig. 2F), while those in the untreated mice did not express these markers. It has been reported that ICAM-1 and VCAM-1 expression is essential for normalization of tumor vasculature, an effect should be important for vaccine efficacy (26, 27). Thus, the closely aggregated structures composed of CD11c⁺ DCs and antigen-specific CD8⁺ T cells around the CD31⁺VCAM-1⁺ICAM-1⁺ vessels seen in tumor sites resembled previously described TLS.

We analyzed antigen-specific CD8⁺ T cells in spleen and tumor sites in tumor-bearing aAVC-treated or untreated mice. As shown in Fig. 3A and B, the frequency of OVA-specific CD8⁺ T cells in the tumor was higher than in the spleen (30% vs. 7% respectively) in aAVC-OVA–treated mice. The OVA-tetramer⁺CD8⁺ T cells in the tumor from aAVC-OVA–treated mice were not only increased in number but also had better function, as measured by expression of IFNγ and CD107 as a degranulation marker in response to OVA antigen (Fig. 3C). These results suggest that the CTLs were either primed by the DCs in lymphoid tissues and subsequently recruited into the tumor, or by the tumor-associated DCs in situ. To better understand the nature of the T-cell response to tumor antigen (OVA), we assessed the TCR repertoire of the OVA-tetramer⁺CD8⁺ T cells of spleen and tumor-infiltrating lymphocytes (TIL) from three tumor-bearing, aAVC-OVA–treated mice (Fig. 3D; Supplementary Table S3). On the basis of the results of deep sequencing, the TCRβ CDR3 repertoire in each of three groups of mice was different, even though all of the cells analyzed were OVA-tetramer⁺CD8⁺ T. For example, the sequence of CDR3 in cls-1 in spleen 1 was different from those of cls-1 in spleen 2 or 3 (Supplementary Table S3). On the other hand, when we compared the TCRβ CDR3 repertoire in the spleen and tumor in the same mouse, it was coincident (Supplementary Table S3). Thus, these TCR repertoire analyses indicate that the dominant OVA-tetramer⁺CD8⁺ T clones that expanded were different in each mouse, but were the same in spleen and tumor of individual mice. CTLs could first be primed by DCs in lymphoid tissues and then subsequently recruited to the tumor sites and further expanded by tumor-associated DCs.

It has been noted in several studies that immunosuppressive molecules are often expressed on TIL (28) and inhibit their function. When TILs in untreated tumor-bearing mice were examined, we detected very few OVA-tetramer⁺CD8⁺ T cells, too few to analyze, but the OVA-tetramer⁻ CD8⁺ T cells expressed relatively high levels of PD-1, and Lag3 (Fig. 4A, bottom). In contrast, OVA-tetramer⁺CD8⁺ T cells in aAVC-OVA–treated mice were plentiful and they expressed low levels of PD-1 and Lag3 (Fig. 4A upper).

Immunotherapy with anti-PD-1 or PD-L1 mAb has shown clinical efficacy with some tumor types (29). We therefore assessed whether there would be any synergistic effect with combined aAVC-OVA and anti-PD-1 mAb therapy but none was observed (Fig. 4B), possibly because the aAVC-OVA therapy was already so effective. We thus reduced the number of aAVC-OVA to 1/50, that is, 1 × 10⁴ cells/mouse, thinking that the antitumor efficacy of the combination of aAVC-OVA and anti-PD-1 mAb would be more apparent under this suboptimal condition. Indeed, T-cell responses in these vaccinated mice were decreased by about 75% (the frequency of OVA-tetramer⁺CD8⁺ T cells in total spleen was % 2.12 ± 0.61 with the normal aAVC-OVA dose vs. % 0.61 ± 0.10 with the lower dose; Supplementary Fig. S5). Although the antitumor effect with aAVC therapy alone was also weakened, we detected a potent synergistic effect with the aAVC-OVA plus anti-PD-1 mAb combination (Fig. 4C). There was no increase in the number of T cells after administration of the anti-PD-1 mAb alone; however, the combination of anti-PD-1 mAb and aAVC-OVA therapy led to an increased frequency of OVA-tetramer⁺CD8⁺ T cells in the tumor as well as in spleen (Fig. 4D). Therefore, this combination therapy would show a therapeutic advantage.

Memory T cells are readily generated after an acute infection; however, it has been difficult to generate memory T cells against cancer cell–associated antigens. If any, most of the studies have had some success by using the transfer of transgenic tumor-specific T cells, but not under physiologic conditions. Therefore, we assessed the induction of memory T cells following aAVC-OVA therapy, without transferring transgenic T cells. As shown in Fig. 5A (top), we detected memory T cells systemically, not only in lymphoid tissues but also in non-lymphoid tissues one year after immunization with aAVC-OVA. Further analysis showed that CD44^hiCD62L^hi central memory T cells (T_CM) are predominant in lymphoid tissues, that is, spleen and lymph nodes, whereas in the lung the T_CM and CD44^hiCD62L^lo T effector memory (T_EM) cells were equivalent. On the other hand, T_EM cells were predominant in the bone marrow (Fig. 5A lower) and liver (not shown). We then assessed the function of the memory T cells in the spleen and bone marrow. They produced IFNγ and TNFα after a stimulation with OVA peptide (Fig. 5B and C), suggesting that multifunctional memory T cells were induced by aAVC-OVA therapy.

Next, we examined whether the long-lived memory T cells could be boosted. If the response to aAVC-OVA was predominantly mediated by allogeneic T cells reacting with NIH3T3 cells, then it likely could not be boosted. However, when we administered aAVC-OVA to mice that had been vaccinated with aAVC-OVA as a homologous prime-boost strategy 6 months previously, a strong secondary response was elicited (Fig. 5D, middle and E). Moreover, peptide-pulsed DCs could also induce a robust T-cell response as a heterologous prime-boost setting (Fig. 5D lower and E). Thus, apparently memory T cells were induced by aAVC-OVA therapy.

OVA-expressing tumor cells have been a very useful model to study tumor immunity, however, OVA is an artificial tumor antigen. To evaluate the T-cell response to an authentic tumor antigen, we established aAVC expressing TRP-2 (aAVC-TRP-2), a melanoma tumor antigen, instead of OVA, and then assessed the antitumor and T-cell responses. We analyzed the T-cell response by IFNγ ELISPOT assay 7 days after administration of aAVC-TRP-2. As shown in Fig. 6A, the TRP-2–specific T-cell response was characterized by antigen-specific IFNγ production. We then assessed the antitumor response in a therapeutic model. B16 melanoma cells were allowed to grow subcutaneously for 7 days and then the mice were treated with aAVC-TRP-2 (Fig. 6B). Compared with untreated mice, the aAVC-TRP-2–treated mice showed significant protection against the melanoma.

Wilms' tumor gene WT1 encodes a protein that has been reported to be a tumor antigen in a variety of malignancies (30, 31). To evaluate whether WT1 mRNA could be used in our vaccine models, we established α-GalCer-loaded, mCD1d mRNA, and WT1 mRNA cotransfected NIH3T3 cells, denoted mouse aAVC-WT1. As aAVC-WT1 vaccine efficacy, we confirmed WT1-specific T-cell responses by IFNγ ELISPOT assay 7 days after administration of aAVC-WT1 (Fig. 6C). Next, we injected mice with a WT1-expressing plasmacytoma cell line (J558-WT1) and then treated them with aAVC-WT1 at day 7. In the control untreated group, all of the mice died within 30 days after inoculation with tumor cells. In contrast, 75% of the aAVC-WT1–treated mice survived more than 3 months (Fig. 6D, middle). However, aAVC-WT1 therapy did not protect mice from the parental J558 or the WEHI3B leukemia cell line (Fig. 6D, left and right), demonstrating specificity. These results clearly indicated that the WT1-specific antitumor effect was generated by aAVC-WT1 treatment. When the surviving mice were rechallenged with the J558-WT1 cells more than 4 months later, all of them were protected (Fig. 6E), but all of them died when injected with irrelevant WEHI3B cells, demonstrating the specificity of the protection (Fig. 6E). The data suggest that aAVC therapy can induce a tumor antigen–specific memory response.

Next, we asked whether human NKT cells can respond to human aAVC-WT1. For this purpose, we established α-GalCer–loaded, hCD1d mRNA and WT1 mRNA–cotransfected HEK293 cells as human aAVC (denoted human aAVC-WT1). At first, we verified that CD1d and WT1 are expressed by HEK293 cells using flow cytometry and Western blot analysis, respectively (Fig. 7A and B). Human aAVC-WT1 contained 178.2 ± 41.3 μg WT1 protein/1 × 10⁶ cells. Human NKT cells recognized and killed α-GalCer–loaded, CD1d-expressing HEK293 but not unloaded CD1d-HEK293 (Fig. 7C). Then, we examined whether NKT and T-cell responses could be induced after coculturing PBMCs and human aAVC-WT1 loaded with or without α-GalCer. As shown in Fig. 7D, NKT cells expressed IFNγ after a 6-hour culture with human aAVC-WT1, whereas they did not produce IFNγ in response to unmanipulated HEK293 cells (Fig. 7D). Also, when we cultured HLA-A24⁺ PBMCs derived from healthy volunteers with human aAVC-WT1, WT1-tetramer⁺ T cells were generated, but no T cells specific for the control HIV peptide were observed (Fig. 7E). The human aAVC-WT1 were more effective at generating WT1-tetramer⁺ CD8⁺ T cells than culture of PBMCs in the presence of WTI peptide and IL2. Some human DCs and monocytes as DC precursor express HLA-DR and/or CD14 (32). To confirm the importance of DCs in this system, we depleted CD14⁺ and DR⁺ cells before the coculture, and no WT1-tetramer⁺ CD8⁺ T cells were generated (Fig. 7F). Thus, aAVC-WT1 can generate human WT1-specific T cells, and this process is dependent on antigen-presenting cells, mainly DCs.

In some types of human cancer, including melanoma (33), colorectal cancer (34), breast cancer (35) and lung cancer (36), a correlation between a low density of CD8⁺ T cells in the tumor and poor prognosis has been reported. These cases where there is minimal T-cell infiltration are termed non-T-cell–inflamed tumors (37) and therefore inefficient T-cell migration into the tumor appears to be a major limitation of cancer immunotherapy (27). We showed how aAVC can show better T-cell–mediated antitumor efficacy at the tumor microenvironment in an aggressive tumor model. As a mechanistic summary, in vivo DC activation can induce the formation of TLS in which DCs recruited from lymphoid tissues may induce the formation of ICAM-1- and VCAM-1–expressing blood vessels, resulting in normalization of the vasculature within the tumor. Subsequently, CXCR3-expressing CTLs are efficiently recruited to CXCR3 ligand-expressing DCs, resulting in the aggregation of CTLs in the tumor. We directly showed such a mechanism for aAVC therapy in this study. There are some fragmentary clinical reports describing a correlation between the existence of tumor-associated TLS in patient samples and clinical prognosis (8). Coexistence of effector T cells and a high density of mature DCs in TLS is correlated with longer term survival in lung cancer patients (38). The immunologic results obtained in murine studies with our vaccine were similar to those in the patients with a good clinical prognosis.

PD-1 blockade is a promising therapy because it can have a beneficial effect on approximately 30% of cancer patients (39); however, some types of cancer patients are refractory to this treatment. There are some reports accounting for the differences between the two patient groups, such as PD-L1 expression on tumors, IFNγ production, and preexisting T cells in the tumor sites (37, 39). Preexisting PD-1^hiCD8⁺ T cells in the tumor site have been known to be dysfunctional, but PD-1^loCD8⁺ T cells are capable of producing IFNγ and TNFα, that is functional (39, 40). In addition, PD-1^loCD8⁺ T cells can easily be stimulated by anti-PD-1 mAb (40). Recently, many strategies have been explored that might be combined with the PD1 blockade therapy. In fact, we showed here that the MO4 melanoma was resistant to anti-PD-1 mAb alone, but that a combination therapy with low dose of aAVC therapy and anti-PD-1 mAb was effective. Infiltrating PD-1^loCD8⁺ T cells in the tumor that were generated by aAVC treatment (Fig. 4A) is likely one reason for this successful outcome. Such an increase of antigen-specific T cells in the tumor by synergistic combination therapy might be useful for patients who are resistant to PD-1 blockade therapy.

Generation of memory T cells is one hallmark of vaccine success that may possibly control the long-term antitumor immune response. Criteria and quality of ideal T-cell memory following vaccination has been defined in many studies, typically in viral infection models, and include: (i) long-term maintenance, (ii) recall response to reencounter with the same pathogen, and (iii) multifunctionality, for example, in terms of cytokine production (41–43). In this study, we verified that the memory T cells induced by the aAVC vaccine completely met the above criteria for memory T cells.

Altogether, our findings with the current aAVC vaccine emphasize a novel role for DC activation in situ that shapes both local (i.e., within the tumor) and systemic (i.e., memory T cells) immune responses due to the linkage of innate and adaptive immunity. Further analysis of these local and systemic immune responses may give us important clues about how the antitumor immunologic network operates and should be helpful for developing the next generation of immunotherapy.

No potential conflicts of interest were disclosed.

Conception and design: K. Shimizu, S.-I. Fujii

Development of methodology: K. Shimizu, M. Asakura, S.-I. Fujii

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): K. Shimizu, S. Yamasaki, J. Shinga, Y. Sato, O. Ohara, H. Yagita, S.-I. Fujii

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): K. Shimizu, T. Watanabe, O. Ohara, M. Asakura, S.-I. Fujii

Writing, review, and/or revision of the manuscript: K. Shimizu, H. Yagita, S.-I. Fujii

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): K. Kuzushima, Y. Komuro

Study supervision: S.-I. Fujii

The authors thank Drs. P.D. Burrows (University of Alabama) and I. Mellman (Genentech) for critical reading of the manuscript and Ms. C. Oikawa, Ms. M. Sakurai, and Mr. M. Kawamura (RIKEN, IMS) for providing technical assistance.

This work is supported by JSPS KAKENHI grants (K. Shimizu and S. Fujii) and the Japan Agency for Medical Research and Development (Translational Research Network Program; S. Fujii).

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.