Arginase II Expressed in Cancer-Associated Fibroblasts Indicates Tissue Hypoxia and Predicts Poor Outcome in Patients with Pancreatic Cancer

An adequate level of arginine in the tissue microenvironment is essential for T cell activity and survival. Arginine levels are regulated by the arginine-catabolizing enzyme, arginase (ARG). It has been reported that arginase II (ARG2), one of two ARGs, is aberrantly expressed in prostate cancer cells, which convert arginine into ornithine, resulting in a lack of arginine that weakens tumor-infiltrating lymphocytes and renders them dysfunctional. However, immune suppression mediated by ARG2-expressing cancer cells in lung cancer has not been observed. Here we studied the expression of ARG2 in pancreatic ductal carcinoma (PDC) tissue clinicopathologically by examining over 200 cases of PDC. In contrast to prostate cancer, ARG2 expression was rarely demonstrated in PDC cells by immunohistochemistry, and instead ARG2 was characteristically expressed in α-smooth muscle actin-positive cancer-associated fibroblasts (CAFs), especially those located within and around necrotic areas in PDC. The presence of ARG2-expressing CAFs was closely correlated with shorter overall survival (OS; P  = 0.003) and disease-free survival (DFS; P  = 0.0006). Multivariate Cox regression analysis showed that the presence of ARG2-expressing CAFs in PDC tissue was an independent predictor of poorer OS (hazard ratio [HR]  = 1.582, P  = 0.007) and DFS (HR  = 1.715, P  = 0.001) in PDC patients. In addition to the characteristic distribution of ARG2-expressing CAFs, such CAFs co-expressed carbonic anhydrase IX, SLC2A1, or HIF-1α, markers of hypoxia, in PDC tissue. Furthermore, in vitro experiments revealed that cultured fibroblasts extracted from PDC tissue expressed the ARG2 transcript after exposure to hypoxia, which had arginase activity. These results indicate that cancer cell-mediated immune suppression through ARG2 expression is not a general event and that the presence of ARG2-expressing CAFs is an indicator of poor prognosis, as well as hypoxia, in PDC tissue.


Introduction
The tumor microenvironment plays important roles in the biological behavior of any tumor, which includes the host immune response, tissue oxygen tension, and cancer-associated fibroblasts (CAFs) [1].
Adequate levels of arginine in the extracellular milieu are crucial for T cell proliferation and activity. [2,3] Arginine is one of the semi-essential amino acids and arginine levels are regulated by arginase (ARG), [4] which hydrolyzes arginine to ornithine and urea. There are two isozymes of ARG, ARG1 and ARG2. ARG1, a cytoplasmic enzyme, is expressed mainly in the liver and detoxifies ammonia. ARG2 is expressed as a mitochondrial protein in a variety of tissues, such as kidney, prostate, and small intestine. Arginine also serves as a substrate for nitric oxide synthase (NOS), yielding nitric oxide (NO) and other reactive nitrogen intermediates.
It has been reported that ARG2 is aberrantly expressed in prostate cancer cells, being involved in tumor immune escape mediated by arginine consumption, resulting in a lack of arginine that weakens tumor-infiltrating lymphocytes and renders them dysfunctional. [5] Prostate cancer concomitantly expresses NOS2, thereby reducing arginine progressively and forming peroxynitrite that triggers T cell apoptosis by inhibiting the signal transduction necessary for cellular activation. [5] However, these immunosuppressive effects through the ARG2 expression in cancer cells were not evident in the next lung cancer study. More than 80% (99/120 cases) of lung cancers expressed ARG2 to variable degrees, although the expression of ARG2 had no effect on clinicopathological characteristics, including the host tumor immune response. [6] Now it is controversial, the impact of ARG2 on the clinical features of human cancers.
Pancreatic cancer [pancreatic ductal carcinoma (PDC)] is the fourth and fifth leading cause of cancer-related death in the United States and Japan, respectively. [7,8] The overall 5-year survival rate for patients with pancreatic cancer is 3-5%, [7,9,10,11] in view of its aggressive growth and early metastatic dissemination. The rate of mortality due to this cancer has shown no obvious improvement for decades. The development of predictive biomarkers to assist selection of patient subsets is useful for studies aimed at reducing the mortality of PDC patients, especially in phase clinical studies designed to evaluate various therapeutic approaches [12].
In the present study, we investigated the expression and clinicopathological significance of ARG2 in PDC. We found that only a few PDC cells expressed ARG2, but noticed that ARG2 was expressed in certain stromal cells present in PDC tissue. We also found that the presence of ARG2-expressing stromal cells in PDC tissue was a clinicopathologically significant variable, being associated with a poorer outcome, as well as an indicator of hypoxia.

ARG2 Expression is Rare in PDC Cells, but Characteristically Expressed in Spindle-shaped Stromal Cells within and Around Necrotic Areas in PDC Tissue
Immunohistochemical analysis revealed that ARG2 expression was present in a small number of PDC cases, where it was expressed focally in PDC cells. In contrast, ARG2 was expressed in spindle-shaped stromal cells within and around necrotic areas ( Figure 1A). ARG2 was stained with a dot-like or coarse granular pattern in the cytoplasm of the spindle cells, compatible with the fact that ARG2 is localized in mitochondria ( Figure 1B). ARG2 was also expressed in Langerhans islet cells and ganglion cells in pancreas tissue under non-pathological conditions. Expression of ARG1 was detected only in the cytoplasm of neutrophils by immunohistochemistry, and not in cancer cells ( Figure 1C).

Prognostic Significance of ARG2 Expression in Stromal Cells
Survival analysis demonstrated an association between the presence of ARG2-expressing stromal cells and shorter OS (P = 0.003) and DFS (P = 0.0006) (Figure 2), although no association was found between the presence of ARG2-expressing cancer cells and any patient survival parameter. When the mean number of positive cells was zero, 1 to 80, and more than 80, the staining grades assigned were zero (absence), one (lower expression), and 2 (higher expression), respectively. As ARG2 expression increased, survival became significantly shorter in terms of both OS and DFS ( Figure 2 Correlation between the Presence of ARG2-expressing Stromal Cells and Other Clinicopathological Variables Table 3 lists the clinicopathological features of patients with PDC. When these features were analyzed for correlations, the presence of ARG2-expressing stromal cells was found to be more likely in cases with poorer tumor differentiation in terms of histological grade, presence of necrosis, and presence of stromal cells expressing CAIX and SLC2A1 (alternatively known as glucose transporter type 1, GLUT1), which are markers of hypoxia. In addition, the presence of ARG2-expressing stromal cells was closely correlated with higher tumor-infiltrating CD68 + macrophages and CD66b + neutrophils, and lower tumor-infiltrating CD4 + T cells and CD8 + T cells. No significant correlation was found between the presence of ARG2-expressing cancer cells and any of these clinicopathological parameters.

Majority of ARG2-expressing Stromal Cells are Cancerassociated Fibroblasts (CAFs) in a Hypoxic State
To determine what kind of stromal cells expressed ARG2, we performed double immunohistochemistry and triple immunofluorescence. ARG2-expressing spindle cells were often positive for a-SMA, vimentin, and collagen type I, rarely positive for CD31 and CD68 (data not shown), and negative for desmin, cytokeratins, and D2-40 ( Figure 3), indicating that the majority of ARG2-expressing stromal cells were CAFs and a minority were endothelial cells or macrophages. However, most of the CAFs in PDC tissue did not express ARG2, except for those present within and around necrotic and myxoid degenerative areas. Surprisingly, ARG2 was not apparently expressed in the stromal or epithelial cells surrounding necrotic tissue or myxoid degenerative areas in intraductal papillary-mucinous neoplasms of the pancreas and necrotic tissue in peptic ulcers (data not shown).
Since ARG2-expressing CAFs were present within and around necrotic areas, we next examined the relationship between ARG2 expression and hypoxia in cancer tissue using double immunostaining. Expression of CAIX, SLC2A1, or HIF-1a was observed in CAFs around areas of necrosis. ARG2 expression was found in these CAIX-, SLC2A1-, or HIF-1a-expressing CAFs themselves, or in CAFs located next to CAIX-, SLC2A1-, or HIF-1aexpressing cells (Figure 4). ARG2 expression in cancer cells was not restricted to areas around necrosis or near CAIX-, SLC2A1-, or HIF-1a-expressing cells (data not shown). These findings suggested that expression of ARG2 was induced in the CAFs under hypoxic conditions, and also that there was no apparent correlation between hypoxia and the expression of ARG2 in cancer cells.

Hypoxia Induces Expression of ARG2 in CAFs Extracted from PDC Tissues
To confirm if expression of the ARG2 gene is induced by hypoxia in CAFs, we carried out in vitro experiments using CAFs extracted from PDC tissues. Induction of ARG2 gene expression by hypoxic stress was found in CAFs at both the transcription and protein levels, and the induced ARG2 had arginase activity ( Figures 5A-C). Expression of ARG2 continued under hypoxic conditions, and the induction of ARG2 by hypoxic stress was reversible. Re-oxygenation decreased the up-regulated expression of the ARG2 gene to a level comparable to that in CAFs cultured under normoxic conditions ( Figure 5A). Hypoxic stress induced accumulation of HIF-1a in CAFs ( Figure 5D). According to the gene database, the 59 flanking region of the first exon of the ARG2 gene has potential binding sites for HIF-1 (59-RCGTG-39) [13]: ACGTG at -425 to -421 and GCGTG at -234 to -230. These findings suggest that expression of ARG2 is induced via the HIF-1 signaling pathway.
Expression of the ARG1, NOS1, NOS2, and NOS3 genes was not significantly induced by hypoxia ( Figures 5A). Expression of the ARG1 and NOS3 genes seemed to be induced by hypoxia, although their levels of expression were very low, almost 1/1000 to 1/10000 of those in a normal tissue, which is one of the major tissues that express the gene. Since all of these genes encode enzyme   molecules, it is unlikely that such small amounts of induced transcripts would be biologically important. Rather, it is speculated that small amounts of contaminating cells such as endothelial cells might have responded to the hypoxic conditions. Peroxynitrite is a reactive nitrogen intermediate produced when both NOS and ARG are present. Since these cells did not express nitrotyrosine examined by immunohistochemistry (data not shown) that is generated by the nitration of tyrosine residues by peroxynitrite, NOS2 expression was not induced in CAFs within and around necrotic areas in PDC tissue. These results suggest that NOS2 is not induced in ARG2-expressing CAFs.

ARG2-expressing CAFs Potentially Affect the Immune Reaction
The normal physiological concentration of L-arginine in serum is around 100 mM (50-150 mM). We determined that 12-15 mM L-arginine would be required for T cell proliferation induced by T cell receptor stimulation under the experimental conditions we employed ( Figures 6A and 6B). Next, we tried to examine if ARG2 induced by exposure to hypoxia affects the proliferation of T cells in vitro. In contrary to the expectation, the conditioned medium that had been used for culturing CAFs under hypoxic conditions did not suppress T cell proliferation significantly in comparison to medium that had been used for culturing CAFs under normoxic conditions ( Figure 6A and B). The discrepancy of the findings to our results of the induction of ARG2 protein with a certain enzymatic activity in CAFs by hypoxic exposure might be caused by the secretion of undetermined molecules that can support T cell proliferation from the CAFs.
In order to determine whether CD3 + T cells are proliferating around ARG2-expressing CAFs in PDC tissue, we performed double immunohistochemistry for CD3 and Ki-67 and compared tumor-infiltrating CD3 + T cells with their proliferating index in the area around ARG2-expressing CAFs to the area within the tumor except for necrotic tissue. It was surprised that there were few CD3 + T cells around ARG2-expressing CAFs. Both the absolute number of tumor-infiltrating CD3 + T cells ( Figure 6C) and the proportion of Ki-67-positive proliferating cells among the CD3 + T cells ( Figure 6D) observed in the area around ARG2expressing CAFs were significantly lower than those observed in the other area. These findings suggest that the adaptive immune response is suppressed in areas around ARG2-expressing CAFs.
The direct effect of ARG2 induced by exposure to hypoxia in CAFs against cancer cells was investigated using an in vitro system in which CAFs extracted from PDC tissues were co-cultured with pancreatic cancer MiaPaCa-2 cells. The proliferation of cancer cells was not significantly affected by ARG2 induced by hypoxia ( Figure 7A), and oxidative stress-induced apoptosis of cancer cells but CAFs themselves was not prevented by polyamine produced by ARG2 in response to hypoxia ( Figure 7C).

Discussion
ARG plays key roles in regulating most aspects of arginine metabolism in health and disease, as arginine is a precursor for several molecules with basically important biological functions, such as urea, nitric oxide, polyamines, proline, creatine, glutamate, and agmatine. [2,4] Though ARG2 is reportedly involved in tumor immune escape mechanisms, [5] its roles in human cancer remain to be fully elucidated. Here we showed that the majority of PDCs did not express ARG2 and that the presence of ARG2expressing PDCs was not significantly correlated with patient outcome. Instead, the presence of ARG2-expressing CAFs was found to be an independent predictor of poorer OS (HR = 1.582, P = 0.007) and DFS (HR = 1.715, P = 0.001) in PDC patients. The presence of ARG2 CAFs in PDC tissue was closely correlated with higher tumor grade, the presence of histological necrosis, the presence of CAIX-expressing stromal cells, and the presence of SLC2A1-expressing stromal cells in PDC tissue. ARG2-expressing CAFs were localized within and around necrotic areas in PDC  tissue and co-expressed CAIX, SLC2A1, and HIF-1a. Furthermore, in vitro experiments revealed that cultured fibroblasts extracted from PDC tissue expressed the ARG2 after exposure to hypoxia. These results indicate that cancer cell-mediated immune suppression through ARG2 expression is not a general event and that the presence of ARG2-expressing CAFs is an independent prognostic factor for PDC patients, and also reflects tissue hypoxia.
Reduced oxygen tension has been reported to affect ARG activity in several tissues and cell types including lung, brain, endothelial cells, and macrophages, [14,15,16] although the effects on ARGs differ among tissues, cells, and species. In humans, it has been shown that hypoxia induces the expression of ARG2 in pulmonary artery smooth muscle cells (PASMC) [17] and lung microvascular endothelial cells. [18,19] To our knowledge, the effects of hypoxia on ARG expression in cancer tissues or fibroblasts have not been demonstrated previously. Our study together with the study that hypoxia induced expression of ARG2 but not ARG1 in PASMCs suggests that ARG2 is the hypoxiainducible isoform. [17] It is consistent with our findings that accumulation of HIF-1a was found in CAFs under hypoxic conditions and also that the 5' flanking region of the first exon of the ARG2 gene has potential binding sites for HIF-1.
Ornithine generated by ARG can be further metabolized to polyamines and proline, which are critical for cell proliferation, differentiation and tissue repair, thus potentially preventing cell death. [20,21,22] Therefore, induction of ARG2 might be a defensive response of CAFs in PDC tissue under severe hypoxic stress potentially leading to cell death. Indeed, our results showed that ARG2-expressing CAFs ameliorated oxidative stress-induced apoptosis of CAFs themselves ( Figure 7C). This is consistent with that fact that these CAFs showed a high level of ARG and a low level of NOS, indicating extensive generation of polyamines and proline in them. These CAFs can also induce fibrosis, since proline is an essential component of collagen. As it has been observed that necrotic tissue is often remodeled to fibrous tissue as a form of healing response, [23] it is suggested that ARG2 induction is involved in fibrosis in hypoxic areas to facilitate post-necrotic fibrosis. Although it is still unclear if ARG2-expressing CAFs promote tumor progression, no significant effects on the proliferation of cancer cells and the protection of oxidative stress-induced apoptosis of cancer cells were found in our in vitro experiments ( Figures 7A and 7C).
Most of the PDC cells did not express ARG2. Lung cancer expresses ARG2, although immune suppression mediated by ARG2-expressing cancer cells has not been observed. [6] Thus, it is suggested that cancer cell-mediated immune suppression by ARG2 (and NOS2) in prostate cancer [5] is probably tissuedependent, and not a general event in the cancer microenvironment. Our investigation of tumor-infiltrating immune cells in  PDCs showed that the presence of ARG2-expressing CAFs was significantly correlated with higher infiltration of CD68 + macrophages (P = 0.05) and CD66b + neutrophils (P,0.0001), and lower infiltration of CD4 + T cells (P = 0.003) and CD8 + T cells (P = 0.036). In addition, tumor-infiltrating CD3 + T cells around ARG2-expressing CAFs were few and showed less proliferation ( Figure 6). These findings suggest that the presence of ARG2expressing CAFs is closely related to the immunosuppressive microenvironment. This situation can be explained in terms of tissue hypoxia, [24] although the extent to which local expression conditions (Re-oxygenation) was also analyzed. Each data column represents the mean relative expression standardized with 18SrRNA 6 SE for triplicate determinations. The SLC2A1 (alternatively known as glucose transporter type 1 or GLUT1) gene is hypoxia-inducible. Significance value (Student's t test) of P,0.01 (*). Some genes are expressed in CAFs at an extremely lower level than in a normal tissue that is the major tissue of expressing the gene. In such cases, comparison of gene expression is shown in the insets with 1/50 to 1/10000 scales. (B) Western blot analysis (upper column) reveals that ARG2 protein expression is induced in CAFs used in (A) upon exposure to hypoxia. N and H indicate the cells cultured under normoxic and hypoxic conditions, respectively. ARG2 protein induced in CAFs has arginase activity (lower column). One unit of arginase converts 1 mmol of L-arginine to ornithine and urea per minute at pH 9.5 and 37uC. Each data column represents the mean activity 6 SE for triplicate determinations. Significance value (Student's t test) of P,0.05 (*) and P,0.01 (**).  of ARG2 in CAFs contributes to this immunosuppression is still unclear. A hypoxic environment attracts infiltrating innate immune cells, which negatively regulate adaptive immunity, [24] and hypoxia also promotes the microenvironment implicated in regulatory T cell development and function, [24] resulting in a strong impairment of T cell function. Immunosuppression induced by hypoxia and that mediated by ARG partly overlap. Hypoxic murine but not human macrophages express high levels of both ARG1 and NOS2 that lead to T cell suppression, where expression of both enzymes causes peroxynitrite generation, loss of CD3f chain expression, and T cell suppression and apoptosis. [25,26] ARG1 acts mainly through depletion of arginine, which impairs T cell signal transduction and function. [27] Further studies are warranted to clarify whether and how ARG2 functions in immune editing in human cancer tissues.
Besides exerting host immunosuppressive effects, hypoxia can lead to the development of aggressive cancer phenotypes such as cell immortalization, autocrine growth/survival, angiogenesis, invasion/metastasis, and resistance to chemotherapy, through a mechanism mediated mainly by HIF-1a. [28,29,30] In the present study, we demonstrated that the presence of hypoxic foci containing ARG2-expressing CAFs was an independent predictor of poor survival in PDC patients, being consistent with our previous study in which the presence of necrosis was shown to be an independent predictor of poor outcome for such patients. [31] These findings and the therapeutic implications of hypoxia make it a high-priority target for cancer therapy. Recently, HIF-1targeting therapy and anti-angiogenesis therapy have been reported to yield promising anti-cancer effects. [29,30,32] It is suggested that evaluation of ARG2-expressing CAFs would be useful not only for decision-making about postoperative clinical management, but also for stratifying patients for clinical trials aimed at evaluating HIF-1 targeting or anti-angiogenesis therapies. . In order to evaluate the effect of oxidative stress, the percentages of living cells after exposure to oxidative stress were divided by the percentages of living cells cultured under the same conditions before oxidative stress (right column). The ratio of living cells before and after oxidative stress decreased significantly in both MiaPaCa-2 cells and CAFs cultured under hypoxic conditions. Blocking the synthesis of polyamines with DFMO increased significantly the degree of oxidative stress-induced apoptosis in the CAFs. Data represent one of three independent experiments. Significance value (Student's t test) of P,0.05 (*) and P,0.01 (**). doi:10.1371/journal.pone.0055146.g007

Patients and Samples
This study was approved by the Institutional Review Board of the National Cancer Center, Japan (#17-77). Informed consent was obtained from all participants involved in this study (the consent was written) and all clinical investigation was conducted according to the principles expressed in the Declaration of Helsinki. Clinical and pathological data were obtained through a detailed retrospective review of the medical records of all 214 patients with ductal carcinoma of the pancreas who had undergone initial surgical resection between 1990 and 2005 at the National Cancer Center Hospital. None of the patients had received any prior therapy, and all had received standard therapy appropriate for their clinical stages. The operative procedures included 141 pancreatoduodenectomies or pylorus-preserving pancreatoduodenectomies, 56 distal pancreatectomies, and 7 total pancreatectomies. Secondary tumors and post-neoadjuvant cases were excluded. All patients had complete medical records, and had been followed in the tumor registries for survival and outcome. The clinicopathological characteristics of the patients are summarized in Table 3. One hundred thirty patients were male and 84 were female, with a mean age of 63.6 years (range, 27-87 years). Every patient was followed up in the outpatient clinic every 1-3 months during the first postoperative year, and every 6-12 months thereafter. No patient dropped out during follow-up. The median follow-up period after surgery was 18.8 (2.6-201) months for all patients. Forty-nine patients (20.4%) were alive at the census date (June 2009); 143 (66.8%) died of pancreatic cancer, and 22 (10.3%) died of other causes. Post-resection adjuvant therapy information was available for 208 patients, of whom 8 received chemotherapy and radiotherapy, 84 chemotherapy alone, and 2 radiotherapy alone; 113 did not receive any additional therapy.

Pathological Examination
All of the ductal carcinomas were pathologically reexamined and classified according to the World Health Organization (WHO) classification, [9] the International Union against Cancer (UICC) tumor-node-metastasis (TNM) classification, [33] and the Classification of Pancreatic Carcinoma of the Japan Pancreas Society. [34] Surgically resected specimens were fixed in 10% formalin and cut into serial 5-mm-thick slices, horizontally in the pancreas head, and sagittally in the pancreas body and tail. All the sections were stained with hematoxylin and eosin (HE) for pathological examination.

Immunohistochemistry
Immunohistochemistry was performed on formalin-fixed, paraffin-embedded tissue sections using the avidin-biotin complex method as described previously. [35] We used 4-mm-thick sections of representative blocks with antibodies against the following: . For antigen retrieval, tissue sections were autoclaved at 121uC for 10 min in the Target Retrieval Solution (DAKO) for nitrotyrosine or in the buffer (20 mM Tris/HCl 1 mM EDTA, pH 9.0) for the other antigens. For semiquantitative assessments of the immunohistochemical results for ARG2, CAIX, and SLC2A1, we selected 10 randomized fields per tumor at a magnification of 6100, and the number of positive cells in each field was counted. The mean number of positive cells in the top three fields was calculated. Two observers (YI and NH), having no access to the patient data, independently evaluated these positively stained cells, and the final value employed was the average of the two counts they made. When the mean number of positive cells was zero, 1 to 80, and more than 80, the staining grades assigned were zero (absence), one (lower expression), and 2 (higher expression), respectively.

Quantitative Evaluation of Tumor-infiltrating Immune/ inflammatory Cells
After immunohistochemistry, the microscopic images were imported as digital photo files using a NanoZoomer Digital Pathology (NDP) system (Hamamatsu Photonics, Japan), and we selected three areas at low magnification in which the immunolabeled cells had infiltrated into tumor most densely. The selected areas were those in which it had been confirmed at high magnification that invasive proliferation of cancer cells was present. We did not select areas into which infiltrated immune/ inflammatory cells had been recruited as a result of secondary tumor effects, such as pancreatitis, necrosis, ulceration, or mucus flooding of the tissue. Using NDP View at a magnification of 6200, immunolabeled lymphocytes, neutrophils (except intravascular neutrophils), or macrophages were then counted by two independent investigators (YI and NH). These observers were blinded to each other and also not provided with any clinical information on the outcome of the patients. The average counts for each of the three areas made by the two observers were then compared, and when the difference between their counts was less than 20% of the maximum value, the average of the two was used as the final count; if the difference exceeded 20%, the observers discussed the reasons for the difference and performed recounts until the difference became less than 20%. For the survival and correlation analyses, patients were divided into two groups showing high and low cell infiltration, representing values higher and lower than the median for tumor-infiltrating immune/ inflammatory cells.

Extraction of Fibroblasts from PDC Tissue
Small pancreatic tissue blocks (100-150 mg) were obtained during pancreas surgery from patients with PDC. The tissue blocks were cut (1-5 mm 3 ) and seeded in 6-cm 2 culture dishes in the presence of Dulbecco's modified Eagle medium supplemented with 10% fetal calf serum and antibiotics. Tissue blocks were cultured at 37uC in 5% CO 2 under a humidified atmosphere. Eighteen hours after seeding, the culture medium was changed, and 24 hours later, the small tissue blocks were transferred to new culture plates. The fibroblasts grew out in high number and purity from the tissue blocks 1 to 3 days later. The small tissue blocks were removed after 2-3 weeks. After reaching confluence, the monolayers were trypsinized and passaged 1:2. The purity of the cells was assessed on the basis of morphology (most cells were stellate, with long cytoplasmic extensions; some were also spindleshaped) and immunophenotypes (vimentin was expressed in 99.9% of the cells from all cases, a-SMA in 73-93%, cytokeratins in 0.4-0.5%, CD68 in 0.2-3%, desmin in 0%, and CD31 in 0-1%). Cell populations between passages 3 and 6 were used for the study.

Western Blot Analysis
CAFs were lysed in a lysis buffer [1% Triton X-100 (Sigma-Aldrich) and a cocktail of proteinase inhibitors (Roche Diagnostics Corporation) in phosphate-buffered saline, pH 7.4]. Equal amounts of protein samples were separated on 7% or 10% polyacrylamide gels and transferred to Immobilon-P membranes (Millipore). The membranes were immersed in a blocking solution (5% skim milk in TBS-T) for at least 1 hr at 4uC, and then incubated overnight at 4uC with antibodies. After being washed, the membranes were incubated for 30 min at room temperature with peroxidase-conjugated anti-mouse IgG (ab9) 2 fragment (Amersham, Arlington Heights, IL). The antigen was detected with enhanced chemiluminescence Western blotting detection reagents (Amersham) according to the manufacturer's instructions.

Arginase Enzyme Activity
After cultivating under normoxic condition or hypoxic condition for 48 hrs, CAFs were lysed in a lysis buffer [0.4% Triton X-100, 150 mM NaCl, and a cocktail of proteinase inhibitors in 50 mM Tris/HCl, pH 7.6]. Arginase enzyme activities of their lysates were measured by QuantiChrom Arginase Assay Kit (BioAssay Systems, Hayward, CA) according to the manufacturer's instructions.

T Cell Proliferation Assay
Human CD3 + T cells were purified from healthy volunteer peripheral blood by Ficoll density gradient centrifugation, followed by isolation with immunomagnetic beads (Miltenyi Biotec, Bergisch Gladbach, Germany). Purity was .95%, as confirmed by flow cytometry. 1610 5 of the T cells were labeled by 5-(and-6)carboxyfluorescein diacetate succinimidyl ester (CFSE) (Invitrogen) according to the manufacturer's instructions, stimulated with anti-human CD3/CD28 antibody-conjugated beads (Invitrogen) in arginine-free medium [arginine/leucine/lysine-free RPMI-1640 (Sigma) supplemented with 2% human AB serum, 50 mg/l of Lleucine, 40 mg/l of L-lysine/HCl, and penicillin-streptomycinglutamine (Invitrogen)] with or without L-arginine for four days, and their proliferation profiles were analyzed by flow cytometry. The conditioned medium after culture of 1610 5 CAFs in a 48-well dish under normoxic or hypoxic conditions in the arginine-free medium with the indicating L-arginine for four days were used instead of the fresh control medium.

Cancer Cell Proliferation and Apoptosis Assay
Ten thousand CAFs were cultured in a 48-well dish for 48 hrs, then 2610 4 human pancreatic cancer MiaPaCa-2 cells (American Type Culture Collection, Manassas, VA) were seeded on the CAFs and co-culture was carried out for 48 hrs under normoxic or hypoxic conditions. The cells were usually cultured in RPMI-1640 with 10% fetal bovine serum and penicillin-streptomycin-glutamine (Invitrogen) at 37uC in a humidified incubator with 5% CO 2 . It is thought that the biological activity of ARG2 is exerted mainly through polyamines, which are metabolites of arginine produced by ARG2 and several other enzymes, including ornithine decarboxylase (ODC). In order to block the biological function of ARG2, we used an ODC inhibitor, alpha-difluoromethylornithine (DFMO) (Sigma). CAFs were precultured in fresh medium with 2 mM DFMO for two days and then co-cultured with MiaPaCa-2 cells in the same medium under normoxic and hypoxic conditions. In order to measure the numbers of both living and apoptotic cells, the cells were trypsinized, and reacted sequentially with a mixture of two primary antibodies [CD44v5 (VFF-8; 1:200, Bender MedSystems, Vienna, Austria) and Emmprin (poly sc-9752; 1:100, Santa Cruz Biotechnology)], a mixture of two secondary antibodies [biotin-conjugated antimouse IgG antibodies and biotin-conjugated anti-goat IgG antibodies (Vector, Burlingame, CA)] and AlexaFluor488-conjugated streptavidin (Invitrogen). After changing the buffer to Annexin V binding buffer (10 mM HEPES, pH 7.4, supplemented with 140 mM NaCl, 2 mM CaCl 2 , and 1% bovine serum albumin), the cells were reacted with AlexaFluor647-conjugated Annexin V (BioLegend) and propidium iodide (PI, Sigma). Flow cytometry analyses were carried out using a FACSCalibur (BD Biosciences, San Jose, CA) and the data were analyzed using CellQuestPro software (BD Biosciences). In our preliminary tests, MiaPaCa-2 expressed CD44v5 and Emmprin but not CEA (Ab-3;

Statistical Analysis
Comparisons of qualitative variables were performed using the x 2 test or Fisher's exact test. One-way analysis of variance (ANOVA) was used to compare the means of three or more groups. The postoperative disease-free survival (DFS) and overall survival (OS) rates were calculated by the Kaplan-Meier method. Univariate analysis was performed for prognostic factors using the log-rank test. The factors found to be predictive by univariate analysis were subjected to multivariate analysis using the Cox proportional hazards model (backward elimination method). Differences at P,0.05 were considered statistically significant. Statistical analyses were performed with StatView-J 5.0 software package (Abacus Concepts, Berkeley, CA).