| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
Cell and Tumor Biology |
Departments of 1 Radiation Oncology and 2 Biomedical Engineering, Duke University Medical Center, Durham, North Carolina; and 3 Texas Medical Center, Texas Heart Institute, Houston, Texas
Requests for reprints: Mark W. Dewhirst, Department of Radiation Oncology, Duke University Medical Center, Room 201 MSRB, Research Drive, Box 3455 Durham, NC 27710. Phone: 919-684-4180; Fax: 919-684-8718; E-mail: dewhirst{at}radonc.duke.edu.
 |
Abstract |
|---|
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
|
Introduction |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
In order to fully exploit the potential for diagnosis and therapy of cancer that these findings may harbor, individual metabolic phenotypes of tumors need to be characterized more carefully. As a central question, it needs to be clarified whether glucose uptake rates can vary between solid tumors, independent of hypoxia.
To identify such differences, we compared glucose and lactate concentration gradients in subregions of two different solid tumors (R3230Ac mammary carcinoma and FSA rat fibrosarcoma) in Fischer 344 rats, in conjunction with assessment of perfused vasculature, and location and degree of hypoxia. We also measured glucose consumption and lactate production of FSA and R3230Ac cells under normoxia and hypoxia in vitro, in order to validate our findings. Our results show that although both tumor types retain the Pasteur effect, i.e., the basic capacity to switch to anaerobic metabolism under hypoxia, glucose levels in nonhypoxic areas of FSA tumors are clearly decreased versus R3230Ac. Lack of free glucose leads to limited FDG uptake in FSA tumors versus R3230Ac. In vitro studies of FSA and R3230Ac cells suggest that higher aerobic glucose uptake is responsible for the observed finding.
|
Materials and Methods |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
0.1 mm3) from donor rats were transplanted into young female Fischer 344 rats and grown for 4 weeks to 1.5 cm in diameter. Prior to removal of the tumor, animals were anesthetized with a single dose of nembutal (50 mg/kg, i.p.) and placed on a temperature-controlled heating pad to maintain body temperature. Three hours before tumor removal, the hypoxia marker EF5 (2-8-N-[2,2,3,3,3-pentafluoropropyl]acetamide) was given i.v. at a dose of 100 µmol/L, using a 10 mmol/L solution in 0.9% saline (21, 22). Ten minutes prior to tumor removal, the perfusion marker Hoechst 33342 was given i.v. at a dose of 1 mg/kg. In preliminary studies, this dose was found to not affect blood pressure (data not shown). Three hours after injection of EF5, the tumor was excised and snap-frozen on aluminum foil over liquid nitrogen vapor. A blood sample of the animal was taken concomitantly. Part of the liver was also removed and snap-frozen. The animals were then euthanized with a single dose of 0.2 mL euthasol (Diamond Animal Health, Des Moines, IA). All specimens were stored at <–70°C until analysis. Cell culture. R3230Ac and FSA cells were seeded at a concentration of 1 x 106 cells in 5 mL DMEM (Invitrogen, Carlsbad, CA) + 10% fetal bovine serum (Hyclone, Logan, UT) + 5% antibiotics/antimycotics (Invitrogen) into a 6-cm diameter cell culture dish and were allowed to settle for 6 hours. Cells were cultured at 37°C either at 20% O2 in 75% N2 + 5%CO2 or under 0.5% O2 in 95% N2/5% CO2 in a temperature-controlled chamber. For analysis of glucose and lactate levels, 50 µL medium was sampled after 0 and 12 hours, respectively.
Quantification of lactate and glucose. For analysis of lactate and glucose concentrations out of cell culture medium, the medium was spun down in a centrifuge column (Pall, East Hills, NY) and subjected to glucose and lactate analysis in a CMA analyzer (CMA Microdialysis, Solna, Sweden), following the manufacturer's instructions. Blood plasma was isolated from whole blood following centrifugation and analyzed for lactate and glucose concentrations in a CMA microdialysis analyzer. Before cryosectioning the tissue samples, two parallel holes were inserted into the tumor biopsies using a 32-gauge needle, in order to provide internal landmarks for subsequent co-registration of consecutive image slices. For bioluminescence imaging, 20-µm-thick sections were cut from tumor and liver samples in a cryomicrotome, and transferred onto glass coverslides. The sections were allowed to freeze-dry at <–70°C for at least 2 weeks before measurements were made. The bioluminescence reaction was carried out in a temperature-stabilized reaction chamber at 25°C, which was placed under a fluorescence microscope (Zeiss, Thornwood, NY) with a cooled 16-bit CCD camera with photon counting capability, using Andor image-capture software (Andor, South Windsor, CT). The microscope stage was protected from extraneous light using a custom-fit black box. The sections were brought into contact with reaction solutions for either glucose or lactate, which were prepared following earlier published protocols (23–25). Light emission from the enzymatic activity of the luciferase reporter was detected through the bottom of the coverslip. Photon flux was integrated at an overall magnification level of x50, over a time interval of 30 seconds after an incubation time of 10 seconds. The resulting images were calibrated using appropriate standards, consisting of cryostat optimum cutting temperature embedding medium (Sakura Finetek, Torrance, CA) with defined concentrations of the metabolites added, and sectioned at 20 µm thickness in a cryomicrotome. The measuring procedures were identical for standards and tumor sections. Evaluation was done in ImageJ by using image histograms acquired from selected regions of interest. From the histograms, percentiles of the pixel intensity distribution were determined. The 5th percentile of the intensity range was regarded as the minimum value, 50% as the median representative value for the area, and the 95th percentile as the maximum value. Prior to the bioluminescence reaction, a bright-field image was taken of the unstained slide. Rough structures like tissue edges, necrosis, and the iatrogenically introduced tissue holes were visualized and used later for image registration. Data from distinct subregions of each tumor, such as the tumor edge, vital, nonnecrotic tumor, and perinecrotic tissue areas were obtained for all tumors. From each tumor, three slides were imaged for regions of interest, to provide average data for such regions. Following the bioluminescence reaction, the slides were imaged for endogenous Hoechst 33342 fluorescence staining, using a UV light source. This image also served for colocalization purposes. Consecutive slides used for the alternative metabolite or EF5 were also imaged for Hoechst 33342.
Quantification of hypoxia. Slight changes in the published EF5 staining protocol were made to accommodate the use of the EF5 slide to image Hoechst 33342 as well as for H&E staining (19): in particular, the staining time had to be significantly shortened, in order to minimize diffusion of the Hoechst 33342 and to preserve the tissue quality for H&E staining subsequent to the imaging session. Ten-micron-thick sections were prepared from the tumor samples in a cryomicrotome. The sections were fixed in 4% paraformaldehyde for 1 hour, rinsed twice in PBS, and blocked with primary antibody dilution buffer (Biomeda, Foster City, CA) for 1 hour at room temperature. Cy3-labeled anti-EF5 antibody (ELK3-51, obtained from Dr. C.J. Koch, University of Pennsylvania, Philadelphia, PA) was diluted 1:2 in primary antibody dilution buffer (Biomeda), and the sections were stained for 1 hour at room temperature in the dark. After rinsing, the slides were stored in 1% paraformaldehyde/PBS at 4°C until they were imaged, which usually occurred on the same day. For imaging, the slide was covered with PBS and a hemocytometer coverslip was applied, using adhesive tape as a spacer. Imaging was carried out using a computer-guided motorized scanning microscope stage (Marzhauser Wetzlar, Germany), attached to a Zeiss fluorescence microscope with a computer-driven shutter controlling the UV light source (Uniblitz, Rochester, NY). Images were taken automatically at a magnification level of 100x and a shutter time of 0.3 seconds, and stitched later in ImageJ. The slides were first scanned for EF5 immunofluorescence and subsequently for Hoechst 33342, using the appropriate filters, respectively. Day to day variation of the UV light source was taken into account by normalizing light intensity against permanently immobilized nonphotobleaching quantum dot slides at 420 and 580 nm (Evident Technologies, Troy, NY). Image analysis was based on pixel intensity, as described previously (21, 22). Quantification of the degree of hypoxia was done by comparing staining intensity to sections of the same tumor tissue line, obtained from other donor rats that had been incubated for 3 hours at 0.2% oxygen in a temperature- and pressure-stated hypoxic chamber in the presence of EF5. Background staining intensity was determined by competed staining (21). After imaging the slides for EF5 and Hoechst 33342, they were washed with PBS and subsequently stained with H&E.
Image co-registration. Alignment of consecutive images of lactate or glucose and EF5/Hoechst 33342/H&E was done using Photoshop. The primary landmarks used for co-registration were the Hoechst 33342 perfusion staining patterns. Secondary landmarks were used to verify the alignment and included the needle holes, edges of the sections, and patterns of necrosis.
|
Results |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
|
|
|
µmol/g) and lowest in the tumor in all animals. Within the tumor subregions, glucose was highest at the edge (R3230Ac, 5-6.5 µmol/g; FSA, 0.5-1 µmol/g), and lowest in vital tumor subregions (R3230Ac, 1-6 µmol/g; FSA, 0 µmol/g) and perinecrotic regions (R3230Ac, 0.5-2.5 µmol/g; FSA, 0 µmol/g). Glucose concentrations in the blood, tumor edge, vital tumor, and perinecrotic regions of FSA-bearing rats were significantly lower than in R3230Ac tumor–bearing animals, as confirmed by t test (liver, no difference; blood, P < 0.05; tumor edge, P < 0.001; vital tumor, P < 0.001; perinecrotic region, P < 0.05; one-way ANOVA was done among the singular data points of each animal to ensure homogeneity of data within a group; Fig. 3).
|
|
In order to examine the overall influence of hypoxia on the lactate concentration between individual tumors, images stained for EF5 were thresholded to quantify the area fraction over 10% of maximum binding. This area fraction was then compared with the overall lactate concentration in vital tumor regions. A positive correlation was seen indicating that higher levels of hypoxia within individual tumors were associated with higher lactate concentration.
Glucose consumption and lactate production of R3230Ac and FSA cells in vitro. Net changes of glucose and lactate concentrations per hour in the medium are outlined in Fig. 3D. Normoxic glucose consumption trended higher in FSA than in R3230Ac cells (–0.30 ± 0.09 versus –0.21 ± 0.05 mmol/L/hour; Fig. 3; P = 0.1, two-tailed t test). No differences in glucose consumption were observed for either cell line when comparing incubation in room air versus hypoxia (R3230Ac, –0.31 ± 0.1 at room air versus –0.30 ± 0.2 mmol/L/hour under hypoxia; FSA, –0.21 ± 0.05 versus –0.23 ± 0.13 mmol/L/hour; Fig. 3). Lactate production was significantly higher under hypoxia than under normoxia in both cell lines (R3230Ac room air, 0.43 ± 0.08; hypoxia, 0.60 ± 0.12; FSA room air, 0.45 ± 0.02; and hypoxia, 0.68 ± 0.11 mmol/L/hour; Fig. 3). Lactate levels were not different between R3230Ac and FSA cells under either incubation condition.
|
Discussion |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
-ketoacids, such as pyruvate, have a protective effect against oxidative stress. Glucose shortage may thus give way to cell damage and necrosis via pyruvate deprivation (29, 30). The occurrence of necrosis close to the tumor edge, as occasionally seen in both tumor types, may be a long-term consequence of the spatial heterogeneity of oxygen supply.
Because the blood is the major conduit for glucose homeostasis in the mammalian body, all tissues displaying steady-state glucose levels below the blood must be regarded as net glucose consumers. The whole-tissue steady-state glucose level in these tissues is dependent on the discrepancy between the local consumption of underlying cells versus supply rate of nearby vasculature (and, in tissues capable of producing glucose, its production rate). Because this relationship is crucial to understanding the nature of the glucose distribution in tumors, we measured glucose levels in liver and blood of the animals, in addition to imaging the tumor. We showed a gradient of glucose with highest levels in the liver (production), intermediate in the blood (homeostasis), and lowest concentrations in the tumor (consumption; Figs. 2 and 3A). The tumors further displayed their highest glucose concentrations at the well-perfused edge, declining towards the more central regions of the tumor. This gradient likely reflects local heterogeneity of uptake versus supply rates.
However, a difference existed in the glucose concentrations in vital, nonhypoxic versus perinecrotic, and hypoxic regions of R3230Ac versus FSA tumors: R3230Ac tumors had significantly higher glucose values in nonhypoxic, vital areas (Fig. 3A, "V"), as compared with perinecrotic, hypoxic areas ("P", paired t test, P < 0.05). In contrast, glucose concentrations in inner regions of FSA tumors were consistently near zero, regardless of hypoxia being present or absent. We interpret this result as an indication that glucose consumption in R3230Ac tumors is restricted in the presence of oxygen, and elevated in the presence of severe hypoxia, as found in perinecrotic regions. Our interpretations are supported by the finding that R3230Ac tumors show the Pasteur effect in vitro, i.e., lactate production increases under hypoxia. The lack of statistical significance in the changes of glucose consumption under hypoxia in these cells can be explained here by the fact that the observed change in lactate production rates of 0.17 mmol/L/hour should theoretically be paralleled by changes in glucose consumption of as low as –0.085 mmol/L/hour. This effect may be covered by variations in measurements.
The virtual absence of free glucose in all inner regions of FSA tumors on the other hand is an indicator that even under nonhypoxic conditions, these tumors consume glucose at rates that are high enough to decrease the steady-state level of free glucose to zero. Indeed, our results show a trend towards an increased aerobic glucose consumption rate in FSA versus R3230Ac cells in vitro (Fig. 3D). In this situation, cellular glucose consumption in the solid tumor is overcoming the supply. The decreased glucose at the tumor edge and even in the blood of FSA tumor–bearing rats underscores our finding that glucose uptake in FSA tumors is limited by supply, not consumption capacity.
To address the question as to whether the two different metabolic phenotypes of R3230Ac and FSA tumors would create different uptake patterns in FDG-PET, we have compared FDG autoradiograph images from R3230Ac and FSA tumors with respective EF5 staining in adjacent sections. R3230Ac tumors show increased uptake in hypoxic regions, whereas FSA tumors display a more uniform uptake, which may be somewhat stronger at the tumor edge because glucose is more abundant there. Example results confirm this conjecture showing an increase in the FDG signal in hypoxic regions of R3230Ac tumors (Fig. 5). Intriguingly, the projected uniform uptake pattern in FSA is observed, but only in slices from deeper tumor regions (Fig. 5, left column). More superficially cut slices, however, show increased hypoxic uptake of FDGs (Fig. 5, middle). In combination with our findings in vitro, showing that hypoxia leads to increased lactate production in FSA cells, we conclude that the Pasteur effect is basically operating in these cells. However, due to diffusion geometry, only peripheral areas of the tumor may be supplied well enough to enable elevated glucose consumption (and lactate production) rates to occur under hypoxia.
|
Our results with the FSA tumors show that local abundance of glucose, which is related to perfusion, can determine the FDG uptake rate. This is consistent with previous findings showing a more or less strict correlation between tumor blood flow and FDG uptake in many human and experimental tumors (1, 13, 15). Furthermore, previous studies monitoring glucose tissue concentrations in several human tumors showed frequent incidence of very low glucose concentrations, indicating that the situation of supply-limited glucose uptake is frequent in clinical tumors (7, 9). Our results explain these findings based on histology and cell culture. We would stress the fact that in order to determine whether glucose uptake or blood flow is measured by FDG-PET, local glucose abundance, i.e., the steady-state concentration, needs to be determined.
Most lactate is either respired in the well-oxygenated myocardium, taken up by the liver for gluconeogenesis, or used by other tissues (31, 32). In normal blood of humans and rats, lactate concentrations are maintained at a level of 0 to 2 mmol/L, and tissues displaying lactate concentrations above this concentration can be regarded as net producers of lactate. However, it has to be kept in mind that some tumor cells may be able to reuse lactate for respiratory purposes when oxygen is present (33, 34).
On a microregional level, a decline of the lactate concentration was observed from the central tumor regions towards the tumor edge in both R3230Ac and FSA tumors (Figs. 2 and 3). Lactate tended to be higher in perinecrotic areas versus vital tumors, a trend which was observed in both tumor types (Figs. 3 and 4). This trend reached borderline significance only in R3230Ac tumors. The underlying mechanism may be an accumulation of lactate in necrotic areas, which is trapped there due to the high production rates of perinecrotic cell populations. However, the bioluminescence method is inadequate to verify this hypothesis, because liquid-filled areas, as present in the necrotic cavity, are lost during the process of sectioning.
Given that there were such striking differences in glucose concentration in these two tumor lines, it was surprising that both cell lines produced lactate at similar rates in vitro. However, several explanations exist for discrepancies between glucose consumption and lactate production in tumor cells. First, respiration rates may differ between FSA and R3230Ac tumors as a consequence of oncogenic events and/or differentiation status. Second, FSA may be shuttling part of the glucose into alternative pathways, such as the pentose cycle. Third, extracellular lactate may be reintroduced into the cell and back-converted into pyruvate by lactate dehydrogenase and subsequently used for cell respiration. Additional studies are ongoing in our laboratory to explore these mechanisms and will be reported separately.
In order to further identify in vivo correlates for the Pasteur or Warburg effect in solid tumors, we compared the thresholded percentage of tissue area >10% maximal binding intensity of EF5 (as a measure of hypoxia) in several slices with the lactate concentration in the sample. A positive correlation was seen between lactate and hypoxic areas if plotting all samples (Fig. 4C). This finding is further evidence that lactate production is dependent on hypoxia in both tumor types and is confirmed by our results in vitro. The inverse relationship between whole-sample average lactate and glucose concentrations in the vital tumor, as seen among R3230Ac tumors (Fig. 3B), may further be interpreted as an indication that glucose is indeed the major source for lactate in R3230Ac tumors in vivo.
By comparing the microregional glucose and lactate concentrations in liver, blood and hypoxic, nonhypoxic, and peripheral areas of R3230Ac mammary carcinomas and FSA fibrosarcomas of the rat, we were able to identify the Pasteur effect, as defined by increased lactate production under hypoxia, in both FSA and R3230Ac tumors. Both tumors however differ in their glucose metabolism, as glucose demand versus supply is drastically increased in FSA tumors, which is probably due to increased aerobic glucose uptake in FSA cells. Different mechanisms may play a role in this phenomenon: potential candidates capable of selectively influencing glucose uptake that have been shown to play a role in cancer are the glucose transporter Glut1 (1), hexokinase II (35), and phosphofructokinase-2. The latter is part of a normally liver-specific feed forward mechanism for glycolysis that has been shown to play a role in cancer (36, 37). Our study is the first to show such differences comparing different tumors in vivo. Our observations also provide an explanation for the observation that FDG uptake and tumor blood flow are correlated with each other, by demonstrating that glucose demand may overcome supply in solid tumors. We conclude that reliable information on glucose uptake can be provided by FDG-PET only if glucose is locally abundant.
|
Acknowledgments |
|---|
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.
We thank Dr. C.J. Koch, University of Pennsylvania, Philadelphia, for kindly providing anti-EF5 antibody (ELK3-51) for this study; Dr. L. Chen and S. Snyder for assistance with the animal work; and E.R. Ward for technical advice concerning the image analysis. The help of Dr. Gamal Akhabani-Hneid, Duke University Medical Center, Department of Radiology, in performing the autoradiographs, and of Dr. P. Sonveaux in preparing the manuscript is greatly appreciated.
Received 11/ 1/04. Revised 3/ 7/05. Accepted 4/ 6/05.
|
References |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
expression under normoxia. J Biol Chem 2002;277:42919–25.-Ketoacids scavenge H2O2 in vitro and in vivo and reduce menadione-induced DNA injury and cytotoxicity. Am J Physiol 1995;268:C227–36.
This article has been cited by other articles:
|
|
 |
|
  H. Harada, S. Itasaka, S. Kizaka-Kondoh, K. Shibuya, A. Morinibu, K. Shinomiya, and M. Hiraoka The Akt/mTOR Pathway Assures the Synthesis of HIF-1{alpha} Protein in a Glucose- and Reoxygenation-dependent Manner in Irradiated Tumors J. Biol. Chem., February 20, 2009; 284(8): 5332 - 5342. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 R. Wirthner, S. Wrann, K. Balamurugan, R. H. Wenger, and D. P. Stiehl Impaired DNA double-strand break repair contributes to chemoresistance in HIF-1{alpha}-deficient mouse embryonic fibroblasts Carcinogenesis, December 1, 2008; 29(12): 2306 - 2316. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 C. C. Riedl, P. Brader, P. B. Zanzonico, Y. S. Chun, Y. Woo, P. Singh, S. Carlin, B. Wen, C. C. Ling, H. Hricak, et al. Imaging Hypoxia in Orthotopic Rat Liver Tumors with Iodine 124-labeled Iodoazomycin Galactopyranoside PET Radiology, August 1, 2008; 248(2): 561 - 570. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 M. C. Skala, K. M. Riching, A. Gendron-Fitzpatrick, J. Eickhoff, K. W. Eliceiri, J. G. White, and N. Ramanujam In vivo multiphoton microscopy of NADH and FAD redox states, fluorescence lifetimes, and cellular morphology in precancerous epithelia PNAS, December 4, 2007; 104(49): 19494 - 19499. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 R. U. Svensson, J. M. Barnes, O. W. Rokhlin, M. B. Cohen, and M. D. Henry Chemotherapeutic Agents Up-regulate the Cytomegalovirus Promoter: Implications for Bioluminescence Imaging of Tumor Response to Therapy Cancer Res., November 1, 2007; 67(21): 10445 - 10454. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 P. Brader, C. C. Riedl, Y. Woo, V. Ponomarev, P. Zanzonico, B. Wen, S. Cai, H. Hricak, Y. Fong, R. Blasberg, et al. Imaging of hypoxia-driven gene expression in an orthotopic liver tumor model Mol. Cancer Ther., November 1, 2007; 6(11): 2900 - 2908. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 E. Takahashi, T. Takano, Y. Nomura, S. Okano, O. Nakajima, and M. Sato In vivo oxygen imaging using green fluorescent protein Am J Physiol Cell Physiol, October 1, 2006; 291(4): C781 - C787. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 K. R. Laderoute, K. Amin, J. M. Calaoagan, M. Knapp, T. Le, J. Orduna, M. Foretz, and B. Viollet 5'-AMP-Activated Protein Kinase (AMPK) Is Induced by Low-Oxygen and Glucose Deprivation Conditions Found in Solid-Tumor Microenvironments. Mol. Cell. Biol., July 1, 2006; 26(14): 5336 - 5347. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 H. Yuan, T. Schroeder, J. E. Bowsher, L. W. Hedlund, T. Wong, and M. W. Dewhirst Intertumoral Differences in Hypoxia Selectivity of the PET Imaging Agent 64Cu(II)-Diacetyl-Bis(N4-Methylthiosemicarbazone) J. Nucl. Med., June 1, 2006; 47(6): 989 - 998. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| Cancer Research | Clinical Cancer Research |
| Cancer Epidemiology Biomarkers & Prevention | Molecular Cancer Therapeutics |
| Molecular Cancer Research | Cancer Prevention Research |
| Cancer Prevention Journals Portal | Cancer Reviews Online |
| Annual Meeting Education Book | Meeting Abstracts Online |
