Abstract
Tumors commonly harbor multiple genetic alterations, some of which initiate tumorigenesis. Among these, some tumor-specific somatic mutations resulting in mutated protein have the potential to induce antitumor immune responses. To examine the relevance of the latter to immune responses in the tumor and to patient outcomes, we used datasets of whole-exome and RNA sequencing from 97 clear cell renal cell carcinoma (ccRCC) patients to identify neoepitopes predicted to be presented by each patient's autologous HLA molecules. We found that the number of nonsilent or missense mutations did not correlate with patient prognosis. However, combining the number of HLA-restricted neoepitopes with the cell surface expression of HLA or β2-microglobulin(β2M) revealed that an A-neohi/HLA-Ahi or ABC-neohi/β2Mhi phenotype correlated with better clinical outcomes. Higher expression of immune-related genes from CD8 T cells and their effector molecules [CD8A, perforin (PRF1) and granzyme A (GZMA)], however, did not correlate with prognosis. This may have been due to the observed correlation of these genes with the expression of other genes that were associated with immunosuppression in the tumor microenvironment (CTLA-4, PD-1, LAG-3, PD-L1, PD-L2, IDO1, and IL10). This suggested that abundant neoepitopes associated with greater antitumor effector immune responses were counterbalanced by a strongly immunosuppressive microenvironment. Therefore, immunosuppressive molecules should be considered high-priority targets for modulating immune responses in patients with ccRCC. Blockade of these molecular pathways could be combined with immunotherapies targeting neoantigens to achieve synergistic antitumor activity. Cancer Immunol Res; 4(5); 463–71. ©2016 AACR.
Introduction
Immune checkpoint blockade therapy has now unequivocally demonstrated robust and durable clinical responses in patients with previously refractory tumors. The existence of antitumor immunity and its contribution to the treatment of cancer is now widely appreciated (1, 2). It is thought that the generation of CTLs plays a pivotal role in mediating antitumor immunity. Several reports support the notion that amino acid substitutions originating from tumor-specific somatic mutations can be recognized by tumor-specific CTLs (3–10). T cells recognizing such individual tumor-specific mutations will not have been subject to central tolerance induction in the thymus. These mutations are therefore likely to represent highly immunogenic neoantigens which can be targeted for tumor immune rejection (11–13). Recent reports suggest that the number of candidate neoantigens predicted by next-generation sequencing (NGS) and MHC class I binding algorithms is correlated with clinical outcomes in melanoma and lung cancer patients receiving anti-CTLA-4 or anti-PD-1 immune checkpoint blockade antibody therapy (14, 15). These findings were interpreted as indicating that endogenous immune responses against mutation-derived neoantigens were augmented by blocking regulatory mechanisms inhibiting antitumor T-cell function.
Clear cell renal cell carcinoma (ccRCC) appears to be an immune-sensitive tumor as suggested by early attempts at immunomodulating therapies, such as IL2 and/or IFNα (16, 17). The results of recent clinical trials of immune checkpoint blockade therapy in ccRCC suggest that this cancer may also be sensitive to these agents, similar to melanoma or lung cancer, despite its lower levels of somatic mutations (18–20). Therefore, we examined the impact of neoantigens in ccRCC patients and investigated whether the number of candidate neoantigenic T-cell epitopes present in ccRCC correlates with clinical outcomes and immune signatures in the local tumor microenvironment.
Materials and Methods
Patients
ccRCC patients (N = 97) from our previous study (21) were investigated in this cohort. The clinicopathologic characteristics of these patients are shown in Table 1. Most of the patients had received standard therapy with surgery alone or surgery plus cytokines, tyrosine kinase inhibitors, and mTOR inhibitors according to the guidelines in the Japanese Urological Association (22) and the National Comprehensive Cancer Network (NCCN). No patients had received immune checkpoint blockade therapy in this cohort.
. | N . |
---|---|
All cases | 97 |
Sex | |
Male | 75 |
Female | 22 |
Age, years | |
<60 | 39 |
≥60 | 58 |
pT | |
1a | 37 |
1b | 27 |
2 | 11 |
3 | 21 |
4 | 1 |
N | |
0 | 90 |
1 | 3 |
2 | 4 |
M | |
0 | 85 |
1 | 12 |
Fuhrman grade | |
1 | 13 |
2 | 56 |
3 | 21 |
4 | 5 |
Undetermined | 2 |
Clinical outcome | |
Alive | 74 |
Dead | 23 |
. | N . |
---|---|
All cases | 97 |
Sex | |
Male | 75 |
Female | 22 |
Age, years | |
<60 | 39 |
≥60 | 58 |
pT | |
1a | 37 |
1b | 27 |
2 | 11 |
3 | 21 |
4 | 1 |
N | |
0 | 90 |
1 | 3 |
2 | 4 |
M | |
0 | 85 |
1 | 12 |
Fuhrman grade | |
1 | 13 |
2 | 56 |
3 | 21 |
4 | 5 |
Undetermined | 2 |
Clinical outcome | |
Alive | 74 |
Dead | 23 |
Abbreviations: M, metastasis; N, nodes; pT, primary tumor.
Whole-exome and RNA sequencing data
The cohort used in the current study consisted of the 97 cases from our previous study in which HLA typing (97 cases), whole-exome sequencing (106 cases), and RNA sequencing data (RNA-seq; 100 cases) were all available (21). The mean coverage by whole-exome sequencing was 128×, and 89% of targeted regions were covered by ≥20 independent reads. Somatic mutations were detected using the EBCall (Empirical Bayesian mutation Calling) algorithm (23). To validate somatic mutations, PCR-based deep sequencing was performed for randomly selected single-nucleotide variants (SNV) and indels. True-positive rates were 96% (504/525) for SNVs and 96% (55/57) for indels. In RNA sequencing, 130 million reads with 100 base length per samples were obtained of which 105 million reads were mapped on RefSeq gene.
HLA typing
HLA types were assigned from exome sequencing data from normal tissues or peripheral blood mononuclear cells using HLA typing software (Omixon target HLA). Three patients with ambiguous predictions were excluded from the analysis, resulting in the 97 ccRCC patients included here.
MHC class I binding prediction
Mutated peptides derived from missense mutations from exome sequencing data were used for MHC class I binding prediction. The missense mutations in genes with low expression that have FPKM value (fragments per kilobase of transcripts per million fragments sequenced) of less than 1 were eliminated. Long peptides containing the predicted mutation or of wild type were assessed using the Immune Epitope Database and Analysis Resource (http://www.iedb.org/) offline; all possible 9- and 10-mer peptides were selected, each predicted to bind to a specific HLA allele for each patient (24). Ranking of predicted binding scores was by MHC class I epitope binding algorithms (NetMHCpan v2.8). Mutated peptides that had an IC50 value below 500 nmol/L were regarded as candidate neoepitopes.
Analyses of immunologic signatures in the tumor microenvironment
The immunologic signature in the tumor was evaluated by gene expression analysis of infiltrating immune cells (CD8A, CD4, CD19, KLRF1, and CD68), phenotypic markers of T cells (CD27, CD28, 4-1BB, OX40, and GITR), and effector molecules of CD8 T cells (PRF1, GZMA, IFNγ, and TNFα) using FPKM values from the RNA-seq data.
Statistical analysis
Survival times were calculated as the number of days from surgery to death, or the last time the patient was known to be alive, and plotted as Kaplan–Meier survival curves. The log-rank test was used to examine the significance of differences in the survival between groups. Comparison of results was performed by an unpaired, two-tailed Student t test with GraphPad Prism 5 (GraphPad Software, Inc.). A value of P < 0.05 was considered statistically significant.
Results
Somatic mutation number does not correlate with ccRCC prognosis
Neoantigens derived from tumor-specific somatic mutations are now acknowledged as major targets for antitumor immune responses. It has been reported that the number of mutations correlates with prognosis in melanoma and lung cancer patients under checkpoint blockade immunotherapy. To determine whether the same may apply to ccRCC (not on immunomodulatory therapy), we investigated the number of tumor-specific somatic mutations in this tumor type. The clinicopathologic characteristics of these 97 ccRCC patients are summarized in Table 1. These patients received standard treatments; TNM staging and Fuhrman grade were strong prognostic values for these patients (Supplementary Fig. S1). The number of nonsilent mutations in these patients ranged from 7 to 155 (median, 50). Most of them were missense mutations (5–130, median, 41; Table 2 and Supplementary Table S1). We divided the patients into two groups with values above or below the median and plotted Kaplan–Meier survival curves. Survival between groups with a higher or lower number of nonsilent mutations was not significantly different (Fig. 1A). The same analysis was true for the missense mutations as well (Fig. 1B). These results document that the number of mutations did not correlate with ccRCC prognosis.
. | Median value . | Range . | N . |
---|---|---|---|
Nonsilent mutations (total) . | 50 . | 7–155 . | 4,859 . |
Missense | 41 | 5–130 | 3,983 |
Indel | 5 | 0–13 | 521 |
Splice | 1 | 0–5 | 126 |
Nonsense | 2 | 0–8 | 221 |
Read-through changes | 0 | 0–1 | 8 |
Median value | Range | N | |
HLA-Restricted neoepitopes (ABC-neo) | 54 | 2–139 | 5,186 |
HLA-A (A-neo) | 20 | 0–88 | 2,198 |
HLA-B (B-neo) | 11 | 0–47 | 1,226 |
HLA-C (C-neo) | 14 | 0–69 | 1,762 |
. | Median value . | Range . | N . |
---|---|---|---|
Nonsilent mutations (total) . | 50 . | 7–155 . | 4,859 . |
Missense | 41 | 5–130 | 3,983 |
Indel | 5 | 0–13 | 521 |
Splice | 1 | 0–5 | 126 |
Nonsense | 2 | 0–8 | 221 |
Read-through changes | 0 | 0–1 | 8 |
Median value | Range | N | |
HLA-Restricted neoepitopes (ABC-neo) | 54 | 2–139 | 5,186 |
HLA-A (A-neo) | 20 | 0–88 | 2,198 |
HLA-B (B-neo) | 11 | 0–47 | 1,226 |
HLA-C (C-neo) | 14 | 0–69 | 1,762 |
Neoepitope number does not correlate with ccRCC prognosis
We determined the most likely neoepitopes among the tumor-specific mutations in each patient based on MHC class I binding prediction scores according to NetMHCpan v2.8. Peptides predicted to bind to each patient's HLA molecules with high-affinity (IC50 < 500 nmol/L) were considered to be likely neoepitopes. Because missense mutations represent the majority of the nonsilent mutations, we focused on neoepitopes derived from these mutations. Although HLA class I–restricted epitopes range from 8 up to 11 or more amino acids, 9- and 10-mer peptides were examined in this study because they account for 90% of the neoepitopes which induce T-cell responses in humans (25). The number of HLA-A–restricted neoepitopes (A-neo) determined by the estimated binding affinity ranged from 0 to 88 (median, 20). Similarly, 0 to 47 (median, 11) HLA-B–restricted neoepitopes (B-neo), and 0 to 69 (median, 14) HLA-C–restricted (C-neo) neoepitopes were identified (Table 2 and Supplementary Tables S2 and S3). Then, the total number of neoepitopes presented by HLA-A, -B, and -C molecules (ABC-neo) ranged from 2 to 139 (median, 54). We divided these patients into two groups according to higher or lower predicted numbers of neoepitopes, designated ABC-neohi (≥54 neoepitopes) and ABC-neolo (with <54). The ABC-neohi group did experience better clinical outcome compared with the ABC-neolo group, but this difference did not achieve statistical significance (log-rank test, P = 0.2413; Fig. 1C). Similar results for correlations with survival were obtained for HLA-A–restricted as well as for HLA-A, -B, and -C–restricted neoepitopes (ABC-neo) in that patients in the A-neohi group (defined as ≥20 neoepitopes) tended to have a better prognosis than those in the low group (<20 neoepitopes), again without achieving statistical significance (log-rank test, P = 0.096; Fig. 1D). In contrast, there was no difference at all between B-neohi (≥11) and B-neolo (<11) groups or between C-neohi (≥14) and C-neolo (<14) groups (Fig. 1E and F).
Neoepitope antigen presentation correlates with ccRCC prognosis
It is widely accepted that tumors can evade antitumor immune responses by the downregulation or loss of expression of the MHC class I molecules responsible for antigen presentation (26, 27). We therefore evaluated the expression of HLA-A, HLA-B, HLA-C, and β2-microglobulin (β2M) using RNA-seq data (21). The FPKM value of HLA-A, HLA-B, HLA-C, and β2M expression was 22,427–216,795 (median, 79,336), 18,298–309,293 (median, 93,251), 18,808–180,437 (median, 53,744), and 52,111–709,571 (median, 253,455), respectively (Supplementary Table S2). Patients whose tumors had high expression of HLA-A, HLA-B, HLA-C, or β2M tended to have a better prognosis than those with low expression, but again this difference did not achieve statistical significance (P = 0.0966, P = 0.2177, P = 0.2308, and P = 0.0797, respectively; Fig. 2A–D). Next, we examined these patients in terms of the number of neoepitopes and the levels of HLA or β2M expression. For this, patients were divided into four groups: (i) neohi/HLAhi; (ii) neohi/HLAlo; (iii) neolo/HLAhi; and (iv) neolo/HLAlo. As shown in Fig. 2E, patients with A-neohi/HLA-Ahi had longer overall survival (OS) than those with A-neolo/HLA-Alo and this time statistical significance was achieved (P = 0.027). However, there was no statistical significance between B-neohi/HLA-Bhi and B-neolo/HLA-Blo or between C-neohi/HLA-Chi and C-neolo/HLA-Clo (Fig. 2F and G). As for β2M expression, ABC-neo was combined with β2M. Patients with ABC-neohi/β2Mhi also had significantly better OS than those with A-neolo/β2Mlo (P = 0.0423; Fig. 2H). These data suggest that antigen presentation of neoepitopes correlates with a better prognosis for ccRCC patients.
Neoepitope presentation determines immune signatures
Antigen presentation of neoepitopes should result in recognition by tumor-specific T cells and the induction of antitumor immune responses in the host and their presence within the tumor. Therefore, we examined the expression of markers of infiltrating immune cells such as CD8A, CD4, CD19, KLRF1, and CD68, as well as T-cell costimulatory molecules such as CD27, CD28, 4-1BB, OX40, and GITR, and finally, effector molecules such as perforin (PRF1), granzyme A (GZMA), IFNγ, and TNFα in the tumor tissue using RNA-seq data (Fig. 3). As mentioned above, patients were divided into 4 groups in terms of the number of potential A-neoepitopes present and the level of HLA-A expression. As expected, CD8A expression in patients with A-neohi/HLA-Ahi was significantly higher than in A-neolo/HLA-Alo (P = 0.0309; Fig. 3A). The expression of other immune cell markers in the four groups was similar. However, patients in the A-neohi/HLA-Ahi group had significantly higher expression of the cytotoxic effector molecules PRF1 and GZMA than in the A-neolo/HLA-Alo group (P = 0.0058 and P = 0.0094, respectively). Similarly, CD8A expression in patients in the ABC-neohi/β2Mhi group was significantly higher than in the ABC-neolo/β2Mlo group (P = 0.0238; Fig. 3B). GZMA expression in ABC-neohi/β2Mhi patients was also significantly higher in the ABC-neolo/β2Mlo group (P = 0.0317). Taken together, these results indicate that presentation of A-neo with high HLA-A or ABC-neo with high β2M expression correlates with an intratumoral immune signature of CD8 T cells and their effector molecules.
Expression of effector immune signature genes does not correlate with prognosis
Next, we examined whether the expression of these immune-related genes had any impact on prognosis. Again, we divided the patients into low and high groups according to the expression of each of these genes but failed to find any correlation with survival (Fig. 4). None of these genes by itself was correlated with prognosis in ccRCC patients. Notably, even expression of genes related to CD8 T cells and their effector molecules, such as CD8A, PRF1, or GZMA was not associated with better prognosis, despite their expression being associated with neoepitope load and antigen presentation.
Positive and negative immune response signatures coexist
As noted above, the number of A-neo or ABC-neo present and higher expression of HLA-A or β2M tended to correlate with better prognosis and with immune signatures of effector CD8 T cells and molecules. However, the expression of CD8 T-cell genes and effector molecules such as CD8A, PRF1, and GZMA did not correlate with better prognosis. Therefore, we divided these ccRCC patients into two groups according to their higher- or lower-than median expression of CD8 genes, PRF1, or GZMA and examined the expression of genes associated with negative regulation of immune responses. As shown in Fig. 5, the immune checkpoint molecules CTLA-4, PD-1, and LAG-3, but not Tim-3, were highly expressed in patients with a CD8Ahi pattern relative to CD8Alo patients (Fig. 5A; P < 0.0001). PD-L1, PD-L2, and IDO1, which are induced by IFNγ, were also more highly expressed in CD8Ahi than CD8Alo patients (Fig. 5A). The same was true for IL10 (Fig. 5A), whereas NOS1 and VEGFA were expressed at equal levels in all patients. Similarly, the expression of these negative regulatory genes (CTLA-4, PD-1, LAG-3, PD-L1, PD-L2, IDO1, and IL10), was associated with PRF1 (Fig. 5B) and GZMA (Fig. 5C). These results indicate that the immunosuppressive microenvironment appears to be dominant in ccRCC tumors.
Discussion
In this study, we analyzed data from whole-exome and RNA sequencing in tumors from 97 ccRCC patients and found that the number of mutations and potential neoantigens derived from them was not associated with prognosis. However, when the number of neoantigens and the expression of HLA-A or β2M were considered together, patients grouped as A-neohi/HLA-Ahi or ABC-neohi/β2Mhi had better clinical outcomes and higher expression of immune-related CD8 T-cell genes and their effector molecules. Nonetheless, the latter were not themselves correlated with prognosis, possibly due to coexpression of genes that were associated with immunosuppression in the tumor microenvironment.
It has been reported that mutational rates differ between different cancers (28). Melanomas have the highest mutational rate (median, 13.2 mutations per Mb) with non–small cell lung cancers (NSCLC) having the second highest rate (median, 8.17 and 6.43 mutations per Mb for the squamous and nonsquamous subtype, respectively; ref. 28). Rizvi and colleagues reported that in 34 lung cancer patients, each had between 11 and as many as 1,192 missense mutations (median, 201) resulting in a predicted range of 8–610 neoepitopes (median, 112; ref. 15). It was also reported that the mutation rate and neoepitope load were indications for sensitivity to immune checkpoint blockade therapy and correlated with longer OS in melanoma and lung cancer patients (14, 15).
In ccRCC patients, the mutation rate was reported as medium to low (median, 1.53 mutations per Mb; ref. 28). Consistent with this, in our 97 ccRCC patients, each had 5–130 missense mutations (median, 41) and 2–139 predicted neoepitopes (ABC-neo; median, 54; Table 2). Unlike results in melanoma and lung cancer patients, the mutational and neoepitope load was not correlated with prognosis in the ccRCC patients in the current study. This might be due to the small number of mutations and low neoepitope load in ccRCC patients. Alternatively, it might be because our ccRCC patients had never received immune checkpoint inhibitors such as anti-PD-1 antibody. In fact, most of the patients reported here had received standard therapy with surgery (nephrectomy) alone or surgery plus cytokines, tyrosine kinase inhibitors, and mTOR inhibitors according to evidence-based clinical practice guidelines for renal cell carcinoma provided by the Japanese Urological Association (22) and the NCCN Guidelines. Whether high mutation/neoepitope load shows any positive association with prognosis in ccRCC patients who do receive immune checkpoint blockade therapy remains to be investigated, as in the melanoma and lung cancer patients mentioned above.
Nevertheless, we did find that ccRCC patients with a high potential neoepitope load and higher-than-median HLA or β2M expression tended to have a better prognosis, coupled with an immunologic signature of cytotoxic T-cell response in the tumor, as assessed by their expression of CD8, PRF1, and GZMA (Figs. 2 and 3). Therefore, these results indicate that intact neoantigen presentation machinery, but not mutation/neoantigen load alone, may contribute to a more favorable prognosis associated with the presence of an effector immunologic signature of CD8 T cells and molecules.
However, we found that the expression of CD8A, PRF1, or GZMA did not correlate with prognosis in ccRCC patients (Fig. 4). Consistent with our results, Nakano and colleagues performed IHC and demonstrated that the infiltration of CD4 and CD8 lymphocytes was correlated with poor survival in ccRCC patients, rather than better survival, as was expected (29). Increased CD8 T-cell infiltration was also associated with poor prognosis and disease progression after IFN therapy (30). Although at first sight paradoxical, these findings may be explained by the fact that immune signatures of CD8 T cells and their effector molecules coexisted with the expression of immunosuppressive molecules in the tumor (Fig. 5). These results are also consistent with previous reports that the immunosuppressive ligand PD-L1 and the enzyme IDO are induced by the IFNγ produced by infiltrating T cells (31, 32). In fact, a statistically significant correlation between IFNγ and PD-L1 expression or the presence of IDO in the tumor was observed (Supplementary Fig. S2).
Neoantigens are potential targets of antitumor immune responses constitutively as well as under immunotherapeutic conditions (33, 34). However, antitumor immune responses may themselves induce counter-regulatory mechanisms to inhibit them as part of physiologic feedback control of immune reactivity. Indeed, immune checkpoint molecules (CTLA-4, PD-1, LAG-3) and their ligands (PD-L1 and PD-L2) as well as IDO and IL10 are upregulated concordantly with higher CD8, PFR1, and GZMA gene expression. Therefore, strategies to terminate the induction of such adaptive resistance (35) and overcome these immunosuppressive microenvironmental factors are required to improve the treatment of ccRCC. These considerations may explain why immune checkpoint blockade is effective in only a fraction of ccRCC patients (20%–30%; refs. 18–20).
Recently, novel strategies to actively induce immune responses targeting neoantigens have been developed (36–38). However, as we demonstrated here, the adaptive resistance operating in ccRCC might also hamper the response to a vaccination targeting neoantigens. Thus, for the induction of synergistic antitumor activity, blockade of these immunosuppressive molecules combined with cancer vaccines against neoantigens could be an ideal strategy in ccRCC patients.
The current study has some limitations. It was based on the prediction of candidate neoepitopes by computer algorithms, as in other earlier reports (39–42). Antigen presentation as reflected in the values of A-neo/HLA-A or ABC-neo/β2M correlated with better prognosis of ccRCC patients, although B-neo/HLA-B and C-neo/HLA-C did not. Whether these results are attributable to the dominant role of HLA-A and neoepitopes restricted by it in immune responses in ccRCC patients or whether there is a problem with the accuracy of prediction of HLA-B- or HLA-C–restricted neoepitopes remains unclear (24). Future studies need to determine whether in silico–predicted neoantigens do indeed induce T-cell responses in ccRCC patients and whether these are different among HLA types. At any rate, the integrated analysis of genetic and immunologic landscape using NGS technologies becomes the main stream for the development of effective cancer immunotherapies (41, 42).
Although ccRCC is accepted as an immune-sensitive cancer, only a very limited number of candidate tumor-associated antigens recognized by CTLs have been identified so far (43–47). Shared tumor antigens, such as cancer-testis antigens, are also expressed in a limited manner (48, 49). In a recent report (41), the expression of several endogenous retrovirus (ERV) genes was positively correlated with immune cytolytic activity in ccRCC patients, suggesting that ERV is likely to be involved in antitumor immunity in these patients. However, the associations between ERV-derived epitopes and prognosis, or between cytolytic activity and prognosis in ccRCC patients, have not been determined. Here we demonstrated that neoantigen load, antigen presentation machinery, and immune signatures in the tumor do influence the prognosis of patients with ccRCC. We suggest that neoantigens might be major targets of choice for manipulating immune responses to control ccRCC, as in melanoma and lung cancer.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
Authors' Contributions
Conception and design: H. Matsushita, K. Kakimi
Development of methodology: K. Kakimi
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): H. Matsushita, Y. Sato, H. Kume, S. Ogawa, Y. Homma, K. Kakimi
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): H. Matsushita, Y. Sato, T. Karasaki, T. Nakagawa, K. Kakimi
Writing, review, and/or revision of the manuscript: H. Matsushita, Y. Sato, T. Nakagawa, H. Kume, Y. Homma, K. Kakimi
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Y. Sato, H. Kume, S. Ogawa, Y. Homma, K. Kakimi
Study supervision: Y. Homma, K. Kakimi
Acknowledgments
The authors thank Ms. Manami Miyai (Department of Immunotherapeutics, The University of Tokyo Hospital and Medinet Co. Ltd.) for excellent technical assistance.
Grant Support
Part of this study was performed as a research program of the Project for Development of Innovative research on Cancer Therapeutics (P-Direct), Ministry of Education, Culture, Sports, Science and Technology of Japan (to K. Kakimi); supported in part also by a Grant-in-Aid for Scientific Research of the Ministry of Education, Culture, Sports, Science and Technology of Japan (to H. Matsushita and K. Kakimi) and the Princess Takamatsu Cancer Research Fund (13-24522; to H. Matsushita).
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.