Prevalence of FOXP3⁺ Regulatory T Cells Increases During the Progression of Pancreatic Ductal Adenocarcinoma and Its Premalignant Lesions

Tumors express many neoantigens originated from the vast number of genetic and epigenetic changes associated with carcinogenesis. Patients with cancer can develop tumor-specific immune responses, although established cancer usually progresses despite the antitumor immune response. It has been suggested that progressive tumors may develop immune escape strategies that include mechanisms to resist immune surveillance and induce immunotolerance. Such mechanisms might include direct deletion of immune effector cells by expression of death-inducing ligands, suppression of tumor-reactive T cells by regulatory T cells (T_R), and tolerization of host T cells by cross-presentation of tumor-derived antigens (1, 2). In contrast to established cancers, there is strong evidence from murine models that specific immune surveillance systems operate at early stages of tumorigenesis, that is, host immune system inhibits the development of tumors (3). It remains to be elucidated how and when effective immune surveillance is overcome during tumor progression. We need effective immunotherapies for aggressive cancers, such as pancreatic cancer, because they are resistant to current treatment. Thus, it is important to understand the mechanisms by which the host immune response changes and the tumor escapes from host immune surveillance during tumorigenesis and tumor progression.

Pancreatic cancer is the fourth and fifth leading cause of cancer-related death in the United States and Japan, respectively (4, 5). Because of its aggressive growth and early metastatic dissemination, the overall 5-year survival rate for pancreatic cancer is <5%, and only 20% of patients can be treated by surgery at the time of diagnosis (6–9). Many morphologic and genetic studies (10–13) strongly suggested that pancreatic intraepithelial neoplasias (PanIN) are the most major premalignant lesions for ductal adenocarcinoma. The next major is intraductal papillary-mucinous neoplasms (IPMN). PanINs are microscopic lesions that are classified into PanIN-1A, PanIN-1B, PanIN-2, and PanIN-3. PanIN-1A is the earliest and mildest change and PanIN-3 is the highest grade lesion, corresponding to carcinoma in situ. IPMNs are a well-characterized clinical and pathologic entity featuring intraductal proliferation of neoplastic mucinous cells, which usually form papillae and lead to cystic dilation of the pancreatic ducts, forming clinically and macroscopically detectable masses (10, 12, 13). Similar to the well-defined adenoma-carcinoma sequence in colorectal cancer and PanINs (14), IPMNs seem to progress from intraductal papillary-mucinous adenoma (IPMA) to IPMN with moderate dysplasia (or borderline IPMN) and then to intraductal papillary-mucinous carcinoma (IPMC) without invasion and eventually to invasive adenocarcinoma.

Recently, thymic-derived CD4⁺CD25⁺ T_R seem to be a functionally unique population of T cells. T_R maintain immune homeostasis in immunotolerance and the control of autoimmunity. It is also related to transplantation immunity and tumor immunity. T_R can inhibit immune responses mediated by CD4⁺CD25⁻ and CD8⁺ T cells in vitro by a contact-dependent and cytokine-independent mechanism (15–18), although more recent reports suggest that the immune suppression mechanisms of T_R in vivo are more complex (19–21). Now, FOXP3 (murine counterpart is Foxp3), a member of the forkhead or winged helix family of transcription factors, is thought to be the most reliable marker for T_R. Foxp3/FOXP3 is disrupted in scurfy mouse (22) and in the human disorder of immune dysregulation, polyendocrinopathy, enteropathy, and X-linked inheritance (IPEX), both of which are characterized by lack of CD4⁺CD25⁺ T_R (23). Foxp3 gene knockout mice show the same phenotype as scurfy mutant mice. The expression of Foxp3 is highly restricted to T_R and is associated with T_R activity and phenotype (24, 25). In fact, gene transfer of Foxp3 converts naive CD4⁺CD25⁻ T cells to functional T_R (26). Therefore, Foxp3/FOXP3 seems to be critical for the development and function of T_R in mouse and human (20, 21, 27). Furthermore, recent reports showed CD4⁺CD25⁺Foxp3⁻ T cells did not show T_R function, although both CD4⁺CD25⁺Foxp3⁺ and CD4⁺CD25⁻Foxp3⁺ T cells had T_R function (28). Therefore, Foxp3/FOXP3 should be used as a marker to evaluate real T_R.

In murine models, it has been described that T_R inhibit the antitumor immune response (19, 29–31). The involvement of CD4⁺CD25⁺ T_R in human cancer has been observed in peripheral blood and tumor tissues from patients with several types of cancer (32–35). In most of the previous studies, T_R cells were collected from fresh tissue, ascites, or blood and analyzed as CD4⁺CD25⁺ cells by flow cytometry. Until the T_R are evaluated using FOXP3-specific antibodies, it has been difficult to analyze the detailed distribution of T_R in tumors and surrounding tissues or to examine the prevalence of T_R in small premalignant lesions detected only by microscopy.

In this study, we established monoclonal antibodies that are specific for human FOXP3 protein and that can be used in immunohistochemistry of paraffin-embedded tissue sections. Using the monoclonal antibodies, we evaluated the prevalence of T_R infiltration in tumor tissues, nonneoplastic inflammatory tissues, and draining lymph nodes in 198 patients with pancreatic ductal adenocarcinoma and 15 patients with nonneoplastic pancreatic lesions. We also investigated the relationship between the prevalence of T_R and clinicopathologic findings. Furthermore, we investigated the relationship between the prevalence of T_R and multistep carcinogenesis by analyzing 84 lesions of PanINs and 51 cases of IPMNs, in addition to analysis of intraepithelial lymphocytes infiltrated in their intraductal premalignant lesions. We provide evidence that the prevalence of T_R increases during the progression of established cancers as well as of their premalignant lesions. Furthermore, the prevalence of T_R was significantly correlated with patient survival, independent of several prognostic factors.

Patients and samples. This study was approved by the Ethics Committee of the National Cancer Center (Tokyo, Japan). Clinical and pathologic data and the specimens used for immunohistochemical analysis were obtained through a detailed retrospective review of the medical records of all 198 patients with pancreatic ductal adenocarcinoma, 51 patients with IPMNs, and 15 patients with nonneoplastic pancreatic lesions who had undergone initial surgical resection between 1990 and 2002 at the National Cancer Center Hospital (Tokyo, Japan). All the patients had not received any prior therapy. All patients with pancreatic ductal adenocarcinoma and with IPMNs received standard therapy appropriate for their clinical stages. Along with tumor extension, lymphadenectomy was done at the hepatoduodenal ligament and around the abdominal aorta. Tumors were classified according to the WHO classification (13, 36), the International Union Against Cancer tumor-node-metastasis classification (37), and the classification of pancreatic carcinoma of the Japan Pancreas Society (38). All patients had complete medical records and had been followed by the tumor registries for survival and outcome. The latest survival data were collected on April 26, 2005. The mean follow-up was 20 months (range, 2-137 months) for patients with pancreatic ductal adenocarcinoma and 45 months (range, 2-142 months) for patients with intraductal papillary-mucinous neoplasia. The clinicopathologic features of the patients are summarized in Table 1.

Establishment of monoclonal anti-human FOXP3 antibodies. The total coding sequence of human FOXP3 amplified by PCR with specific primers (5′-ATgCCCAACCCCAggCCT-3′ and 5′-ggggCCAggTgTAgggTT-3′) was subcloned into the pBAD/Thio-TOPO vector (Invitrogen, Carlsbad, CA). After transformation of Escherichia coli TOP10 with this plasmid vector, the expression of FOXP3-HP-thioredoxin fusion protein was induced with arabinose and the bacterial cells were lysed with a lysis buffer [100 mmol/L NaH₂PO₄, 10 mmol/L Tris/Cl (pH 8.0), 8 mol/L urea]. To purify the FOXP3 fusion protein, the lysate was separated by 7.5% SDS-PAGE followed by copper staining with 0.3 mol/L CuCl₂ solution. The 60-kDa-sized single band representing FOXP3-HP-thioredoxin fusion protein was cut from the polyacrylamide gel and the FOXP3 fusion protein was extracted with a Centrilutor (Millipore, Bedford, MA). The gel extract was concentrated with a Centriprep (Millipore) and dialysed against PBS (pH 7.4). Six-week-old female BALB/c mice (Clea Japan, Inc., Tokyo, Japan) were immunized s.c. with 100 μg of purified FOXP3 fusion protein mixed with complete Freund's adjuvant per mouse. The second to fourth immunizations were done with 50 μg of the purified FOXP3 fusion protein mixed with incomplete Freund's adjuvant at 2-week intervals. After the last injection, immune splenic lymphocytes were fused with SP-2 myeloma cells by using polyethylene glycol (mixture of MW 3,350 and MW 6,000 at 3:1) according to a method described previously (39). Primary screening of the supernatants of hybridomas was done with formalin-fixed paraffin-embedded sections of lymph nodes by immunohistochemistry described below. The supernatants from several different hybridoma clones reacted exclusively with nuclei of lymphoid cells. After repeated cloning, we selected three hybridoma clones, 42, 64, and 243, for further characterization. The immunoglobulin subclass was determined with a mouse monoclonal antibody isotyping kit (Amersham, Buckinghamshire, United Kingdom). These three clones secreted immunoglobulins of subclasses IgG_1κ, IgG_2aκ, and IgG_1κ, respectively.

Expression of FOXP3 protein in mammalian cells and immunocytochemistry. FOXP3 protein was expressed as described previously (40). Briefly, the full-length cDNA of human FOXP3 was subcloned into pcDNA3.1/Hyg (+) (Invitrogen) and the new vector was named pcDNA3.1-FOXP3. Chinese hamster ovary (CHO) cells (American Type Culture Collection, Rockville, MD) cultured on coverslips were transiently transfected with pcDNA3.1-FOXP3 or pcDNA3.1/Hyg by use of LipofectAMINE Plus (Invitrogen). At 48 hours after the transfection, the CHO cells were fixed with 4% paraformaldehyde containing 0.1% Triton X-100 at 4°C for 30 minutes or with chilled methanol with 10% acetone on ice for 20 minutes. The cells were stained by the avidin-biotin complex method of immunohistochemistry described below modified by using FITC-conjugated avidin instead of avidin-biotin complex method reagent and the labeling intensity was observed by confocal microscopy.

Immunoblot analysis and immunoprecipitation. CHO cells were transiently transfected with pcDNA3.1-FOXP3 or pcDNA3.1/Hyg. Forty-eight hours after transfection, both FOXP3 and mock transfectants were lysed in a lysis buffer [1% Triton X-100, a cocktail of proteinase inhibitors (Roche Diagnostics Corp., Indianapolis, IN) in PBS]. Equal amounts of protein samples were separated on 7% or 10% polyacrylamide gels and transferred to polyvinylidene difluoride membrane (Immobilon-P; Millipore). After blocking the membranes in 5% skim milk in 0.05% Tween 20 in TBS, they were incubated with anti-FOXP3 antibody followed by the incubation with peroxidase-conjugated anti-mouse Ig F(ab′)₂ fragment (Amersham). The antigen was detected with enhanced chemiluminescence Western blotting detection reagents (Amersham) according to the manufacturer's instructions. For immunoprecipitation, equal amounts of cell lysates were precleared with protein G-Sepharose 4 Fast Flow (Amersham), then anti-FOXP3 antibody was added to the lysates, and the samples were incubated. Immune complexes were precipitated by incubating the samples with the protein G-Sepharose beads and subjected to immunoblot analysis with anti-FOXP3 antibody as described above.

Immunohistochemical analysis. Immunoperoxidase staining by the avidin-biotin-peroxidase complex method was done on formalin-fixed paraffin-embedded 4-μm-thick sections as described previously (41). To enhance the signal for CD4 (used at a dilution of 1:100; 1F6, Novocastra Laboratories Ltd., Newcastle, United Kingdom) and CD8 (used at a dilution of 1:100; 4B11, Novocastra Laboratories), we used the catalyzed signal amplification (CSA) system (DAKO, Glostrup, Denmark) according to the manufacturer's instructions. No significant staining was observed in the negative controls, which were prepared by using the same class of mouse immunoglobulin at the same concentration.

Serial sections were prepared from each paraffin block. The first section was stained with H&E and the second and third sections were subjected to immunohistochemistry to detect the CD4 and FOXP3 antigens. CD4⁺ or FOXP3⁺ lymphocytes were counted in the corresponding visual field. Quantitative evaluation of lymphocytes was done by analyzing at least three different high-power fields (×40 objective and ×10 eyepiece). We analyzed representative blocks and omitted pinpoint lesions because they were too small to analyze the lymphocytic infiltration at least three different high-power fields. The proportion of FOXP3⁺ lymphocytes in the CD4⁺ lymphocytes was calculated for each field and the averages were compared. Preliminary study revealed that absolute number of lymphocytes infiltrated was easily affected by inflammation. To reduce the effect of inflammation in evaluation of T_R, we selected the prevalence of FOXP3⁺ T_R in CD4⁺ T cells.

Double immunofluorescence. We did double immunofluorescent staining on formalin-fixed paraffin-embedded sections. The 4-μm-thick sections were deparaffinized and autoclaved for 10 minutes at 121°C either in 10 mmol/L citrate buffer (pH 6.0) for CD3 (used at a dilution of 1:50; UCHT1, DAKO) and CD8 or in 1 mmol/L EDTA (pH 8.0) for CD4 and CD25 (used at a dilution of 1:50; ACT-1, DAKO) followed by a wash in 0.05% Tween 20 in TBS (pH 7.6). For double staining of the sections with FOXP3 and CD3, CD4, CD8, or CD25, the last four antigens were stained by the CSA system with our modification and then FOXP3 was stained with CSAII (a biotin-free tyramide signal amplification system from DAKO). The CSA system was modified as follows. After reaction with biotin-conjugated tyramide solution, the sections were incubated with Texas Red–conjugated avidin (1:200; Vector Laboratories, Inc., Burlingame, CA) for 30 minutes at room temperature. To detach the antibodies, the sections were incubated in 100 mmol/L glycine/HCl (pH 2.2) for 2 hours at room temperature with gentle stirring. The sections were then washed in 0.05% Tween 20 in TBS and stained with FOXP3 using CSAII according to the manufacturer's instructions. Just after reaction of the sections with FITC-conjugated tyramide, the sections were washed and mounted with Vectashield mounting medium (Vector Laboratories). For double staining of the sections with FOXP3 and CD20 (used at a dilution of 1:200; L26, DAKO) or CD56 (used at a dilution of 1:100; NCC-Lu-243, Nihonkayaku, Tokyo, Japan), the sections were first immunostained with FOXP3 using CSAII and then stained with CD20 or CD56. Immunostained tissue sections were analyzed with a confocal microscope (LSM5 Pascal; Carl Zeiss Jena GmbH, Jena, Germany) equipped with a 15-mW Kr/Ar laser.

Statistical analysis. Values were expressed as mean ± SD. Statistical analyses were done with StatView-J 5.0 software (Abacus Concepts, Berkeley, CA). Association of the prevalence of FOXP3⁺ cells in CD4⁺ T cells with various clinicopathologic variables was assessed by the χ² test. Kruskal-Wallis nonparametric tests were used to compare the value of numbers of CD8⁺ intraepithelial lymphocytes (IEL)/prevalence of T_R in stromal-infiltrating CD4⁺ T cells among each grades of PanINs and IPMNs. Survival rates were calculated by the Kaplan-Meier method. Differences between survival curves were analyzed by the log-rank test. To assess the correlation between survival time and multiple clinicopathologic variables, multivariate analyses were done by the Cox proportional hazards regression model. Differences were considered significant when P < 0.05.

Establishment of monoclonal antibodies specific for FOXP3 expressed preferentially in CD4⁺CD25⁺ T_R. We established monoclonal antibodies (42, 64, and 243) against FOXP3 protein (Materials and Methods). These antibodies stained the nuclei and sometimes the cytoplasm of CHO cells transfected with FOXP3-pcDNA3.1, which contains cDNA encoding full-length FOXP3 but did not stain mock transfectants (Fig. 1A). When we used these antibodies to do immunoblot analyses of the cell lysates obtained from both transfectants, we detected a condensed band showing the expected size of ∼50 kDa in the lysate from FOXP3-transfected CHO cells. No band was recognized in lanes loaded with lysates of mock transfectants (Fig. 1B). These findings indicate that the monoclonal antibodies specifically recognize FOXP3 protein. Furthermore, the antibodies could immunoprecipitate FOXP3 protein from the diluted lysates of FOXP3-transfected CHO cells (Fig. 1C).

To examine the immunophenotype of FOXP3⁺ cells, double immunofluorescent staining was done. All FOXP3⁺ cells were CD3⁺ (data not shown), almost all of them were CD4⁺ T cells (Fig. 1D), and the majority of the FOXP3⁺ cells were CD25⁺ (Fig. 1F). In contrast, no expression of FOXP3 was found in either CD20⁺ B cells or CD56⁺ natural killer cells (data not shown). The majority of CD8⁺ T cells also did not express FOXP3, although a few did so (Fig. 1E). These findings confirm that the majority of FOXP3⁺ cells are CD4⁺CD25⁺ T cells, with a minor population of FOXP3⁺CD4⁺CD25⁻ T cells and a few FOXP3⁺CD8⁺ T cells. FOXP3⁺ cells were distributed mainly in the paracortical area in lymph nodes (Fig. 1G and H). There were no FOXP3⁺ tissues or cells, except lymphoid cells in brain, liver, lung, kidney, esophagus, stomach, intestine, colon, pancreas, heart, skin, and skeletal muscle. Previous studies have shown that FOXP3⁺CD4⁺CD25⁺ T cells exclusively belong to T_R (18, 20, 21, 27). These findings allowed us to conclude that the monoclonal antibodies we established can specifically detect T_R in formalin-fixed paraffin-embedded tissue sections. We used anti-FOXP3 antibody 42 in the later experiments.

Increased populations of FOXP3⁺T_R in CD4⁺ T cells in tumor stroma of pancreatic ductal adenocarcinoma. Stromal-infiltrated lymphocytes both from tumor stroma and from nonneoplastic inflammatory stroma in patients with pancreatic ductal adenocarcinoma (n = 198) and nonneoplastic pancreatic lesions (n = 11) were examined for the prevalence of FOXP3⁺ T_R. Histologically, lymphocytic infiltration was present in cancer stroma and was usually seen frequently in obvious cancer invasion, particularly in areas where cancer cells had just invaded the surrounding tissue. The prevalence of T_R in CD4⁺ T cells was not markedly different between the central and peripheral areas of cancers. Representative immunohistochemical features are shown in Fig. 2A-D. T_R usually infiltrated into stroma and rarely into the epithelial layer of ducts, although T_R occasionally attached directly to cancer cells. Summarized data from all patients indicated that the prevalence of tumor-infiltrating T_R in CD4⁺ T cells in pancreatic ductal adenocarcinoma (34.6 ± 10.9%) was significantly higher than that in nonneoplastic inflammatory areas from patients with pancreatic ductal adenocarcinoma (7.9 ± 4.9%; P < 0.0001) and in nonneoplastic pancreatic lesions (10.1 ± 4.6%; P < 0.0001; Fig. 2E). The prevalence of T_R in CD4⁺ T cells was significantly higher in less-differentiated adenocarcinoma [well-differentiated adenocarcinoma versus moderately differentiated adenocarcinoma (P = 0.0003); moderately differentiated adenocarcinoma versus poorly differentiated adenocarcinoma (P = 0.0039); Fig. 2E]. Although lymphocytic infiltration was rare in nonneoplastic and noninflammatory areas, it was sometimes marked in nonneoplastic but inflammatory areas compared with cancer stroma. Despite this finding, the prevalence of T_R in CD4⁺ T cells was significantly increased in the stroma of adenocarcinoma compared with nonneoplastic inflammatory areas, suggesting that adenocarcinoma cells recruit T_R to themselves. Our observations also suggest that the prevalence of T_R is closely correlated with cancer progression.

Prevalence of T_R in CD4⁺ T cells and clinicopathologic features of pancreatic ductal adenocarcinoma. We next analyzed the correlation between clinicopathologic features of pancreatic ductal adenocarcinoma and the prevalence of T_R in CD4⁺ T cells. Patients with pancreatic cancers were divided into two groups of high T_R (n = 94) and low T_R (n = 104), representing higher and lower than average (34.6%) prevalence of T_R in CD4⁺ T cells. High prevalence of T_R was significantly correlated with distant metastasis (P = 0.0035; Table 2), advanced tumor stage (P = 0.029), and high tumor grade (P = 0.0002) among the various clinicopathologic features tested (Table 2).

Prognostic significance of prevalence of T_R in CD4⁺ T cells. Overall survival was analyzed in these patients. Fourteen of 198 (7%) patients survived >5 years after surgical resection, and 24 (12%) patients survived >3 years. The low-T_R group showed significantly better survival rates than did the high-T_R group (P < 0.0001, log-rank test; Fig. 2F). Mean survival (±SD) was 26.2 (±29.4) months for the low-T_R group and 13.4 (±15.0) months for the high-T_R group. In the low-T_R group, 21 of 24 (87%) patients survived >3 years after the operation and 15 patients died within a year. In the high-T_R group, 3 of 24 (13%) patients survived >3 years after the operation and 47 patients died within a year. We found that lymph node metastasis, distant metastasis, and tumor margins were independent factors for patient survival in our retrospective study using multivariate analysis for the standard clinicopathologic factors. In multivariate Cox proportional hazard analysis for clinicopathologic variables and prevalence of T_R in CD4⁺ T cells, the hazard ratio for poor prognosis was 2.45 for patients in the high-T_R group compared with patients in the low-T_R group (P < 0.0001; Table 3).

Increased populations of T_R in CD4⁺ T cells in tumor stroma correspond to progression during multistep carcinogenesis both in PanINs to invasive ductal adenocarcinoma and in IPMNs. The prevalence of T_R in CD4⁺ T cells in the ducts of PanINs and IPMNs was analyzed during tumorigenesis and progression of pancreatic adenocarcinoma. Thirty lesions of PanIN-1A (Fig. 3A and B), 30 lesions of PanIN-1B (Fig. 3C and D), 20 lesions of PanIN-2, and 4 lesions of PanIN-3 were randomly selected for this examination. We carefully defined PanIN-3 lesions to avoid possible contamination by intraductal spreading of invasive ductal adenocarcinoma. For this definition, we cut all the pancreas to make sections and carried out histopathologic examination and then defined PanIN-3 as an apparently solitary lesion that was not connected to invasive ductal adenocarcinoma. All these lesions were distant from invasive cancer. However, it is sometimes difficult to differentiate PanIN-3 from intraductal spreading of invasive ductal adenocarcinoma when the intraductal lesion is located next to the invasive ductal adenocarcinoma with obscure connection or without an abrupt transition from markedly atypical to normal-appearing epithelium. For such lesions in our series, we established another category “intraductal spreading of ductal adenocarcinoma without invasive cancer cells surrounding the duct” (Fig. 3E and F) and additionally examined the prevalence of T_R in such lesions. We found a significant increase of the prevalence of T_R in CD4⁺ T cells during the progression from low-grade PanIN to invasive ductal carcinoma [PanIN-1A versus PanIN-1B (P = 0.0004), PanIN-1B versus PanIN-2 (P = 0.0007), PanIN-2 versus PanIN-3 (P = 0.011), PanIN-2 versus intraductal spreading (P = 0.0005), PanIN-3 versus invasive ductal adenocarcinoma (P = 0.012), and intraductal spreading versus invasive ductal adenocarcinoma (P < 0.0001); Fig. 3G]. The prevalence was similar between PanIN-3 and intraductal spreading of adenocarcinoma (P = 0.68).

IPMNs usually contain lesions with various degrees of cellular and structural atypism and are diagnosed as the highest grade of the lesions. To examine the relationship between the prevalence of T_R and each grade of IPMN lesion during tumor progression, we classified IPMNs from 51 cases into 27 lesions of IPMA (Fig. 3H and I), 10 lesions of borderline (Fig. 3J and K), 19 lesions of IPMC without invasion (Fig. 3L and M), and 13 lesions of IPMC with invasion according to the WHO classification. Then, we evaluated FOXP3⁺ and CD4⁺ T cells in each grade of IPMN lesion. As shown in Fig. 3N, the prevalence of T_R in CD4⁺ T cells increased significantly in a stepwise manner and parallel with the progression of IPMNs [IPMA versus IPMN, moderate dysplasia (P = 0.01); IPMN, moderate dysplasia versus IPMC without invasion (P = 0.01)]. There was a particularly marked increase in prevalence when IPMC began to invade (P = 0.0009). These findings suggest that the prevalence of T_R is closely correlated with the progression of premalignant lesions, both PanINs and IPMNs. Clinical follow-up revealed that tumors recurred in 4 of 17 patients with IPMC with invasion, of whom 3 died 17, 21, and 23 months after the operation. No significant difference in the prevalence of T_R was noted between cases with recurrence and no recurrence.

Host immune response was marked in low-grade premalignant lesions but diminished and eventually disappeared during carcinogenesis. To estimate the extent of host immune response to these intraductal lesions, we did immunohistochemistry to detect infiltrating lymphocytes. We found that CD8⁺ T cells infiltrated as IEL in premalignant lesions as well as stromal-infiltrating lymphocytes around the ducts (Fig. 4C and D). In noninflammatory and nonneoplastic ducts, no IELs were apparent (Fig. 4A and B). In both inflammatory and premalignant lesions, the majority of IEL in small- and large-sized pancreatic ducts consisted of CD3⁺CD8⁺ T cells with some CD4⁺ T cells, a few CD56⁺ natural killer cells, and no CD20⁺ B cells. Interestingly, there was marked infiltration of CD8⁺ IEL in PanIN-1A and PanIN-1B, significantly decreased numbers in PanIN-2, and few IEL in PanIN-3 or intraductal spreading of adenocarcinoma [PanIN-1B versus PanIN-2 (P = 0.0055); PanIN-2 versus intraductal spreading (P = 0.011); Fig. 4A-F and J]. Most of these IELs showed cytoplasmic granular expression of TIA-1 (T-cell intracellular antigen-1), indicating that they were cytotoxic T cells (Fig. 4G).

In IPMNs, there was a marked infiltration of IEL in low-grade IPMNs, but infiltration was never or seldom observed in IPMC [IPMA versus IPMN, moderate dysplasia (P = 0.05); IPMN, moderate dysplasia versus IPMC without invasion (P < 0.0001); Fig. 4K]. These findings suggest that a host immune response was present in low-grade premalignant lesions, diminished during the progression of the premalignant lesions, and then almost disappeared in noninvasive adenocarcinomas. It was very rare that FOXP3⁺ T_R infiltrated as IEL in pancreatic ducts and we did not observe CD8⁺ IEL bound to T_R IEL in any stages of lesions (Fig. 4H and I). When we compared the ratio of CD8⁺ IEL counts with the prevalence of FOXP3⁺ T_R infiltrated in stroma around the ducts, the ratio showed the same tendencies to the changes of amounts in CD8⁺ IEL infiltrated during the progression of PanINs (P < 0.0001, Kruskal-Wallis test) and IPMNs (P < 0.0001, Kruskal-Wallis test).

Prevalence of T_R in CD4⁺ T cells in draining lymph nodes was not significantly different among nonneoplastic lesions, premalignant lesions, and invasive adenocarcinoma of pancreas. We examined the prevalence of T_R in CD4⁺ T cells in the draining lymph nodes of the pancreas of 60 patients with invasive ductal carcinoma, 15 patients with nonneoplastic pancreatic lesions, and 51 patients with IPMNs. Patients with IPMNs were classified according to the highest grade of the lesion in the IPMNs. No significant difference in the prevalence of T_R in CD4⁺ T cells was found among lymph nodes from patients with ductal adenocarcinoma, IPMNs, and nonneoplastic pancreatic lesions (Fig. 4L).

In this study, we found that there is a relationship between T_R and pancreatic carcinogenesis. To detect T_R in microscopic premalignant lesions, we raised new monoclonal antibodies to FOXP3. The prevalence of FOXP3⁺CD4⁺ T_R in tissue-infiltrating CD4⁺ T cells was evaluated in pancreatic ductal adenocarcinomas, their premalignant lesions, and nonneoplastic inflammatory lesions. The prevalence of T_R was significantly increased in the stroma of pancreatic invasive ductal carcinomas compared with that in the stroma of nonneoplastic inflammation of the pancreas in patients with pancreatic ductal adenocarcinoma and nonneoplastic pancreatic lesions. The prevalence of T_R was significantly correlated with clinicopathologic factors and patient prognosis. Furthermore, the prevalence of T_R increased during the progression of the invasive cancers as well as of the premalignant lesions of pancreatic ductal carcinoma, PanINs, and IPMNs. It was inversely correlated with the numbers of cytotoxic CD8⁺ T cells infiltrated as IEL in the pancreatic ducts. Thus, we provided the first evidence that cytotoxic T cells and T_R infiltrate during multistep pancreatic carcinogenesis.

Naturally arising CD4⁺CD25⁺ T_R characteristically express CD25, CTL-associated antigen 4 (CTLA-4), glucocorticoid-induced tumor necrosis factor receptor family–related gene (GITR), surface transforming growth factor-β (TGF-β), and FOXP3. CD25 is a critical molecule for proliferation and survival of CD4⁺CD25⁺ T_R (20). However, CD25 is not a suitable marker to define T_R because activated T cells generally express CD25. Compelling studies have revealed that CTLA-4 and TGF-β play roles in the suppressive activity of CD4⁺CD25⁺ T_R against CD4⁺ or CD8⁺ T cells, although they are not expressed exclusively in T_R. Experiments with Foxp3-overexpressing transgenic or Foxp3 gene-depleted mice and other studies have shown that Foxp3/FOXP3 is a master control gene for the development and function of natural CD4⁺CD25⁺ T_R (20, 21, 27, 28). Thus, Foxp3/FOXP3 is thought to be a suitable single marker for detecting CD4⁺CD25⁺ T_R. Our newly established antibodies reacted specifically with FOXP3 protein. Most FOXP3⁺ cells were CD4⁺CD25⁺ T cells and the remaining minor population of FOXP3⁺ cells was CD4⁺CD25⁻ T cells. Zelenay et al. (42) reported that FOXP3⁺CD4⁺CD25⁻ T cells convert to CD25⁺ after homeostatic expansion and/or activation in the murine model and they concluded that FOXP3⁺CD4⁺CD25⁻ T cells constitute a peripheral reservoir of committed regulatory cells. These findings indicate that our antibodies detected the entire population of regulatory CD4⁺ T cells committed to FOXP3⁺CD4⁺CD25⁺ T_R. The prevalence of T_R in our study seemed to be higher than in previous studies (33). We found that the intensity of staining for T_R usually differed between cancer stroma and inflammatory areas, and our values for prevalence are limited to cancer stroma alone, with little dilution of the values by contaminating noncancer stroma.

In murine models, T_R inhibit the antitumor immune response mediated by CD4⁺CD25⁻ T cells and CD8⁺ cytotoxic T cells (19, 29–31). In humans, the prevalence of CD4⁺CD25⁺ T cells in CD4⁺ T cells has been shown to increase in blood, ascites, or cancer tissues of patients with several cancers (32–35); the CD4⁺CD25⁺ T cells from these organs had also suppressive activities against the immune responses. Recent studies, which evaluated T_R as FOXP3⁺ lymphocytes showed that FOXP3⁺ T_R infiltration of the tumor is a significant prognostic factor in the univariate and multivariate analyses in head and neck cancers (43) and Hodgkin's lymphoma (44) and that CD4⁺CD25⁺ T cells contain both FOXP3⁺ T_R and activated lymphocytes infiltrated in cancer stroma. Our study has shown not only that the prevalence of FOXP3⁺ T_R is increased in cancer tissue of pancreatic ductal adenocarcinoma but also that this increase is significantly correlated with tumor progression. Furthermore, our study has shown that premalignant lesions of invasive ductal adenocarcinoma of the pancreas, both PanINs and IPMNs, show an increase in the prevalence of T_R during tumor progression. In contrast, infiltration of CD8⁺ cytotoxic T cells, which were the major lymphocytes to attack neoplastic cells and potential targets for T_R, was prominent in low-grade premalignant lesions, gradually decreased during progression of the tumor, and eventually almost disappeared in PanIN-3, intraductal spreading of adenocarcinoma, and IPMC (Fig. 4J and K). These findings suggest that tumor evasion from host antitumor immunity begins in these premalignant lesions. The details of immune suppression mechanisms of T_R have not been clearly elucidated yet. Initially, the in vitro assays proposed the mechanisms were contact dependent and cytokine independent (15–18). Those in vivo, however, are thought to be more complex (19–21). Some reports suggested that cytokines, such as TGF-β, play important roles in the event (19, 45). Several studies supported the hypothesis that T_R can localize and expand together with effector T cells in antigen-draining lymph nodes. Inflamed tissue facilitates specific suppression that is initiated by antigenic stimulation and local recruitment of T_R together with effector lymphocytes (21). Another studies provided the findings that tumor antigen-specific T_R at tumor sites have a profound effect on the suppression of antitumor immune responses (31, 46). There is opposite correlation between the prevalence of T_R and the numbers of CD8⁺ IEL in premalignant lesions during the pancreatic cancer progression, although no interaction of both cells in epithelial layer was observed. It is consistent with the hypothesis that T_R suppress the CD8⁺ IEL. It also gives rise to the question where T_R affect CD8⁺ lymphocytes becoming IEL and whether T_R inhibit the recruitment of CD8⁺ lymphocytes becoming IEL into epithelial layer of pancreatic ducts, although it is possible that neoplastic cells inhibit the recruitment of cytotoxic T cells independent of T_R function. In the future, we need to make clear how these events happen.

The immune surveillance theory of cancer was proposed in the 1970s but remained controversial for a long period. In the last decade, studies with several gene-disrupted or antibody-treated immunocompromised mouse models have provided evidence that the host immune system controls tumor development (3). Nevertheless, the histologic features of cancer immune surveillance in human tissues remain unclear. If the immune surveillance theory applies to pancreatic carcinogenesis, cytotoxic lymphocytes would attack transformed cells or early-stage tumor cells developing in thin or thick pancreatic ducts. Some transformed cells would disappear, whereas others would be resistant to attack and would remain or progress to higher-grade premalignant cells. Our histologic findings may show the process of immune surveillance in the premalignant lesions of pancreatic ductal adenocarcinoma as well as tumor evasion from the host immune system by immunosuppression mediated by FOXP3⁺CD4⁺ T_R.

Our study showed that the high prevalence of T_R in cancer stroma of pancreatic ductal adenocarcinoma was closely correlated with several malignant features, such as distant metastasis, high tumor grade, and advanced pathologic tumor-node-metastasis stage, suggesting that the prevalence of T_R can be a positive indicator of tumor aggressiveness. Several prognostic factors were examined to determine their effect on survival in patients who underwent resection. In our retrospective study using multivariate analysis for the clinicopathologic variables, lymph node metastasis, distant metastasis, and tumor margins were independent factors for patient survival as reported previously (8, 9). This study showed that the prevalence of T_R in tissue-infiltrating lymphocytes was significantly and negatively correlated with patient survival, independent of tumor-node-metastasis classification, tumor grade, and tumor margins. These findings indicate that the prevalence of T_R is a good predictor of prognosis for patients with ductal adenocarcinoma of the pancreas.

It has been controversial whether the prevalence of T_R in tumor-draining lymph nodes is increased or decreased compared with that in non–tumor-draining lymph nodes (33, 35). Those studies considered CD4⁺CD25⁺ T cells as T_R. In our study, the prevalence of T_R in draining lymph nodes was similar among patients with pancreatic adenocarcinoma, IPMNs, and nonneoplastic lesions. Recently, Chen et al. (19) reported that T_R expand in draining lymph nodes after cognate T-cell receptor stimulation as do naive T cells. More recently, Hiura et al. (47) showed that both T_R and antitumor effector T cells are primed in the same draining lymph nodes of mice with transplanted tumors. They also showed that the CD4⁺CD25⁺ T cells in the draining lymph nodes could be divided into two different populations, CD62L^highCD4⁺CD25⁺ T cells that show T_R activity and express Foxp3 and CD62L^lowCD4⁺CD25⁺ T cells that are effector T cells without T_R activity and Foxp3 expression (47). These findings in murine models imply that T_R in tumor-draining lymph nodes should be expanded, although the prevalence of T_R in CD4⁺ T cells might be determined by the total spectrum of immune response in the tissues from which the draining lymph nodes collect lymph.

In conclusion, our data suggest that T_R play a role in controlling the immune response to pancreatic ductal carcinoma from the premalignant stage to established cancer. In premalignant lesions, both PanINs and IPMNs, host immune surveillance is prominent in the earlier stages but decreases during tumor progression, inversely correlating with the increase in the prevalence of T_R. A high prevalence of T_R seems to be a marker of poor prognosis.

We thank Drs. Yae Kanai, Yuri A. Fukasawa, Hidenori Ojima, and Yoshinori Ino for useful discussions.