Evidence that hypoxia markers detect oxygen gradients in liver: pimonidazole and retrograde perfusion of rat liver.

Nitroimidazole markers of tumour hypoxia bind to normoxic liver and the question has been raised whether this is due to low oxygen concentration or microregional activity of specialised nitroreductases. To answer this question, the binding patterns of the 2-nitroimidazole, pimonidazole, were compared following perfusion of surgically isolated rat livers in anterograde and retrograde directions. Perfusion at low flow rates in anterograde or retrograde directions can be used intentionally to alter oxygen gradients without altering enzyme distributions. Perfusion by means of the portal vein (anterograde direction) produced pimonidazole binding in the pericentral region of liver similar to that observed for pimonidazole binding in vivo. A complete reversal of this binding pattern occurred when the isolated liver was perfused by way of the central vein (retrograde direction). In this case, pimonidazole binding occurred in the periportal region. The extent and intensity of binding in the periportal region during perfusion in the retrograde direction was similar to that in the pericentral region during perfusion in the anterograde direction. It is concluded that low oxygen concentration rather than the non-homogeneous distribution of nitroreductase activity is the primary determinant of 2-nitroimidazole binding in liver. ImagesFigure 3

Nitroimidazoles such as misonidazole bind to hypoxic mammalian cells with an oxygen dependence similar to that for radioresistance. This has led to the use of misonidazole and related compounds as markers of hypoxic, radiation-resistant tumour cells. Numerous investigations have established the usefulness of the approach (Chapman, 1991) but nitroimidazoles also bind to some normoxic tissues and Cobb et al. (1990a,b) have questioned whether hypoxia marker binding is due to oxygen-dependent processes in all cases. A possible alternative is that local distributions of specific nitroreductase activity produce 2-nitroimidazole binding via oxygen-independent pathways. In support of this view, P450dependent enzymes with expected nitroreductase activity (Belisario et al., 1990;Cenas et al., 1994) are known to be located in the pericentral region of liver tissue (review, Jungerman and Katz, 1989) where 2-nitroimidazoles bind Cobb et al., 1990a,b). Furthermore, average oxygen concentrations of 20-120 IM in normal tissues (Jones, 1985;Thurman et al., 1986) might be expected to strongly inhibit 2-nitroimidazole binding given that the binding rate decreases sharply at oxygen concentrations above approximately 14 liM as measured by oxygen microelectrodes (Mueller-Klieser et al., 1991). Nevertheless, there is experimental evidence that 2-nitroimidazole binding in many normal and tumour tissues is primarily dependent on oxygen concentration and not on the presence of specialised nitroreductases in the tissues. For example, MacManus et al. (1989) found that lowered tissue oxygen concentrations created by hypobaric oxygen inhalation increased the levels of misonidazole binding in mouse liver, kidney, spleen, heart and tumour tissues. Van Os-Corby et al. (1987) reported that neither the rates of binding nor the oxygen dependence of binding of misonidazole to isolated liver tissue was significantly different from that for brain, heart and tumour tissues. They also reported that binding rates in isolated hepatocytes were similar to or lower than those for other cells, and it was concluded that bioreductive enzymes in hepatocytes do not confer unique binding properties on these cells (Van Os-Corby, 1986). Parliament et al. (1992) found that inhibitors of the oxygen-independent nitroreductase, NAD(P)H:quinone acceptor oxidoreductase (DT diaphorase), did not significantly inhibit the in vivo binding of misonidazole to mouse liver, parotid gland, oesophageal mucosa or tumour tissue. These results support the idea that oxygen-dependent processes determine marker binding in most normal and tumour tissue but they do not prove that patterns of marker binding in normal tissues are solely due to regions of low oxygen concentration. It will be difficult to resolve the relative importance of microregional nitroreductase activity and local oxygen gradients in many normal tissues, but current knowledge of sublobular compartmentation (Thurman and Kauffman, 1985) in conjunction with the perfused liver model (e.g. Ballet and Thurman, 1991) lays a solid basis for answering the question in liver. As noted above, nitroimidazole binding in liver occurs near the central vein which is a region of low oxygen concentration  and high cytochrome P450-dependent, redox enzyme activity (Jungerman and Katz, 1989). The question is whether elevated P450dependent nitroreductase activity can circumvent oxygen dependence for nitroimidazole binding and, combined with oxygen-independent nitroreductases such as DT diaphorase, weaken the link between tissue P02 and marker binding. One way to answer this question is to vary P02 gradients while leaving nitroreductase distribution unchanged. This can be achieved in isolated livers by perfusion at low-flow rates whereby regions of hypoxia can be created intentionally without changing enzyme distributions (Thurman and Kauffman, 1985). For example, perfusion at low flow rates through the portal vein of liver (anterograde direction) creates hypoxia in the pericentral regions while perfusion of the organ through the central vein (retrograde direction) creates hypoxia in the periportal regions. In order to relate the present study to that of Cobb et al. (1990b), pimonidazole, along with associated immunochemical reagents was used as the hypoxia marker. We have found that perfusion in the retrograde direction produces a complete reversal of the pericentral binding of pimonidazole observed when perfusion is in the anterograde direction. It is concluded, therefore, that P02 is the major determinant of 2-nitroimidazole-binding patterns in liver.

Reagents
Sodium pentobarbitol (Nembutal) was obtained from Aldrich (Milwaukee, WI, USA). Racemic pimonidazole hydrochloride was synthesised in our laboratories according to published procedures (Smithen and Hardy, 1982) and characterised by standard chromatographic, elemental analysis and spectrographic techniques. Radioactive pimonidazole labelled with tritium at the 2-position of the sidechain was prepared in our laboratories by an adaptation of the technique used to label misonidazole (Born and Smith, 1983). The tritiated product was shown to have a radiochemical purity greater than 91% and to co-chromatograph in both thin-layer and high-performance liquid chromatography systems with authentic pimonidazole. Tanks of analysed gas mixtures used to test the oxygen dependence of pimonidazole binding to EMT6 cells were purchased from Matheson Gas Products (Morrow, GA, USA). Goat serum; goat anti-rabbit IgG conjugated to alkaline phosphatase; lipid-free, bovine serum albumin (BSA, product number A-028 1); phenylmethylsulphonyl fluoride (PMSF) and the chromogenic substrate for alkaline phosphatase (Sigma 104 phosphatase substrate) were obtained from Sigma (St Louis, MO, USA). Chemicals used in the enzyme-linked immunosorbent assay (ELISA) and for the preparation of formalin-fixed, paraffin-embedded tissue sections were obtained in reagent grade purity from local suppliers. Reagents for the BCA protein assay were purchased from Pierce Chemical (Rockford, IL, USA). Proteinase K was obtained from Life Technologies (Gaithersburg, MD, USA) and Brij-35, pronase E, Meyer's haematoxylin, crystal mount and buffers used for washing slides during immunostaining procedures were obtained from Biomeda (Foster City, CA, USA). Vector Laboratories (Burlingame, CA, USA) supplied the ABC peroxidase Vectastain kit; avidin -biotin blocking kit, rat adsorbed horse-antimouse antibodies; and, DAB peroxidase substrate. Monoclonal antibody isotyping was carried out with a Clonotyping System/AP kit purchased from Fisher Scientific (Pittsburgh, PA, USA).
Polyclonal and monoclonal antibodies For technical reasons, rabbit polyclonal antibodies were used for the ELISA (Figures 1 and 2) and a mouse monoclonal antibody was used for the immunohistochemical studies (Figures 3 and 4). Polyclonal antisera were used for the ELISA because they had been calibrated previously. Monoclonal antibodies were used for immunohistochemistry because they were more compatible with the batchprocessing, capillary-action, immunostaining technique used for the slide-mounted tissue sections (Microprobe, Fisher Scientific). Competitive ELISA indicated that monoclonal and polyclonal antibodies recognised the same antigen; i.e. the sidechain of pimonidazole.
For the preparation of the immunogens for both polyclonal and monoclonal antibody production, pimonidazole was bound to lipid-free, thiolated BSA by means of a radiation chemical reduction described previously (Raleigh and Koch, 1990). BSA adducts of tritium-labelled pimonidazole of known specific activity were prepared in a similar manner and used to calibrate the ELISA. The solid phase antigen for the ELISA was prepared by radiation chemical reduction of pimonidazole in the presence of thiolated Ficoll. Thiolated Ficoll was produced by an adaptation of published procedures (Inman, 1975) with N-succinimidyl 3-(2-pyridyldithio)propionate (Pierce) serving as the source of latent thiol groups.
The rabbit polyclonal, anti-pimonidazole antisera used in the calibrated ELISA were prepared and characterised in a manner analogous to that for CCI-103F -a 2-nitroimidazole compound with six fluorine atoms on the sidechain (Raleigh et at., Cline et al., 1990). The mouse monoclonal antibody (MAb) to protein-bound pimonidazole was pre-pared by the North Carolina State University Hybridoma Facility (Raleigh, NC, USA). The mouse MAb and rabbit polyclonal antisera were used without further purification.
Isotyping showed the MAb to be IgG, with no detectable contamination with other immunoglobulins.
Analysis of tissue bound andfree pimonidazole Competitive ELISA for pimonidazole binding in lysates of EMT6 cells and in diluted liver homogenates was performed using the calibrated ELISA based on rabbit polyclonal antisera in a manner analogous to that used previously for CCI-103F (Raleigh et al., 1994;Thrall et al., 1994).  Figure 1 Competitive ELISA showing the inhibition of antipimonidazole polyclonal antisera binding to a solid phase, Ficollbased antigen by selected antigens. The hapten concentrations for the peptide adducts derived from reductively activated pimonidazole bound to BSA (pimo-BSA) were calculated from the specific activity of the tritium-labelled pimonidazole used to prepare the BSA adduct. Data points are from a single experiment incorporating replicate ELISA measurements for each hapten with standard deviations typically ± 5%. A decreasing rate of chromogen production indicates an increasing inhibition of antisera binding to the solid phase antigen by the various soluble antigens: BSA-pimonidazole (d); free pimonidazole (d); and free hexafluorinated CCI-103F (A). Antigens showing no cross-reactivity include misonidazole, benznidazole and etanidazole (data not shown). ELISA for which protein-bound [3H]pimonidazole served as the primary standard (see below). Weighed samples of rat liver (100 mg) were minced and suspended in 10 volumes of phosphate-buffered saline/0.05% Tween (PBS-Tween) solution in a 5 ml round-bottomed glass tube. The suspension was homogenised for 10 s at the highest setting in an Omni Mixer fitted with a Minimicro generator (Omni International, Waterbury, CT, USA). One aliquot of the homogenate was taken for protein determination by means of the bicinchoninic acid (BCA) reagent. A second aliquot was analysed by ultraviolet spectroscopy (UV) for the presence of unchanged pimonidazole and a third aliquot was analysed for tissue-bound pimonidazole by ELISA.
For the UV analysis of unchanged pimonidazole, homogenate samples were diluted 1:1 (v/v) with 10% aqueous trichloroacetic acid (TCA). The suspension was centrifuged for 10 min at 10000 r.p.m. in an Eppendorf model 5415 Microfuge (Brinkman Instruments, Westbury, NY, USA). The supernatant was diluted 1:4 in distilled water and analysed for pimonidazole (324 nm; molar extinction coefficient = 7810) by means of a Beckman Model DU 70 UV spectrophotometer (Beckman Instruments, Fullerton, CA, USA). Unlabelled liver tissue treated in this way showed no interfering absorption at 324 nm. The limit of detection for pimonidazole was estimated to be 1 ,iM.
For the ELISA, the homogenates were diluted 1:1 with PBS-Tween containing 1.0 mg ml-' of proteinase K (20 units mg-') and the mixtures were incubated overnight at 37°C in a shaking water bath. The protease inhibitor, PMSF, was added to a final concentration of 400 jiM and the mixtures were heated for 10 min at 95°C in a hot water bath in order to completely inactivate the proteinase K. The sample was centrifuged for 10 min at 10 000 r.p.m. in the Eppendorf Microfuge and aliquots of the supernatant were used for the ELISA. A goat anti-rabbit secondary antibody conjugated to alkaline phosphatase combined with a chromogenic substrate was used to measure the amount of pimonidazole antisera bound to the solid phase antigen in the ELISA plates. ELISA values were measured by means of a Molecular Devices Vmax kinetic plate reader (Molecular Devices, Palo Alto, CA, USA) in terms of the rate of chromogen formation (405 nm). The results were analysed by means of Delta Soft software (BioMetallics, Princeton, NJ, USA) and reported as milli optical density units per min (milli OD min-', Figure 1).
Oxygen dependence ofpimonidazole binding to mammalian cells In order to relate the oxygen dependence of pimonidazole binding to that published for well studied misonidazole , EMT6 cells at 3.1 ± 0.87 x 105 cells ml were suspended in 25 ml of PBS containing 100 jiM pimonidazole hydrochloride. Cell suspensions were placed in a series of silanised (Sigmacote; Sigma) glass, 125 ml gas collection tubes which were thoroughly washed with distilled water and autoclaved before use. The tubes were mounted on the deck of an orbital shaker (Model SS110504, Integrated Separation Systems, Natick, MA, USA) so that cell suspensions could be constantly agitated throughout the gas exchange and incubation phases of the experiment. Cell suspensions were equilibrated with gas phases containing 5% carbon dioxide and varying amounts of oxygen (12 to 2 x 105p.p.m. as analysed by the supplier) in nitrogen. The equilibration was achieved by means of 15 gas exchanges under partial vacuum over a period of 10 min. Cells were then incubated in the presence of pimonidazole for 2 h with agitation under continuous gas flow at 37°C. Control experiments showed that cell viability as measured by trypan blue exclusion was essentially unaffected by this treatment. Cells were harvested, washed extensively to remove unbound pimonidazole and analysed by ELISA for covalently bound pimonidazole and by the BCA reagent for protein content Perfusion of rat livers Female Sprague Dawley rats (120-130 g) were given standard laboratory chow and water ad libitum and then fasted 24 h before the beginning of the perfusion experiments. The animals were anaesthetised with 50 mg kg' intraperitoneal sodium pentobarbitol and their livers (typical weight 3.5-4.0 g) were surgically isolated and perfused in the anterograde direction at a 'normal' flow rate (4 ml min-' g l) at 37°C with haemoglobin-free, Krebs-Henseleit bicarbonate buffer, pH 7.4 (118 mM sodium chloride, 24.9 mM sodium bicarbonate, 1.19 mM potassium dihydrogen phosphate, 1.18 mM magnesium sulphate, 4.74 mM potassium chloride and 1.27 mm calcium chloride) saturated with 95% oxygen and 5% carbon dioxide in a non-recirculating mode. After 20 min of perfusion at 4 ml min-' g-', livers were perfused in either anterograde or retrograde directions for 50 min at low flow rate (1 ml min-' g-') with Krebs-Henseleit buffer containing 413 jiM pimonidazole hydrochloride. This concentration of pimonidazole was used in order to give a strong binding signal during the short perfusion time. The duration of the perfusion was chosen to be within the viability limits of the model as determined previously by trypan blue uptake and lactate dehydrogenase release (Bradford et al., 1986).
Following perfusion, one liver lobe was removed, flash-frozen in liquid nitrogen and stored frozen for subsequent ELISA. The remaining liver tissue was perfusion-fixed with 10% formalin. A sample of fixed tissue was excised, embedded in paraffin and sectioned at 6 jim onto silanised, precoated glass slides for immunohistochemical analysis. Control liver samples unlabelled with pimonidazole were also prepared.
Immunohistochemistry Immunohistochemistry was carried out as described previously for CCI-103F . Sections of formalin-fixed and paraffin-embedded liver tissue were deparaffinised by treatment with xylene, a graded series of alcohol and water mixtures and finally with water. Hydrated tissue sections were treated briefly with 0.01% protease (pronase E) in order to enhance antigen availability, washed and exposed to mouse anti-pimonidazole IgG, MAb in PBS-Tween for 2 h at 37°C. A second antibody comprising ratabsorbed, horse anti-mouse antibody conjugated to peroxidase was then applied to the tissue sections for 30 min. Immunostaining of sublobular regions was achieved by adding DAB peroxidase substrate to the sections followed by incubation at 37°C for 20 min. The immunostained sections were lightly counterstained with haematoxylin and mounted with crystal mount solution.

Image analysis
An Image-1/AT image acquisition and analysis system (Universal Imaging, Chester, PA, USA) incorporating an Axioskop 50 microscope (Carl Zeiss, Thornwood, NY, USA) was used to capture and analyse images of immunostained tissue sections. Ten periportal and ten pericentral fields were chosen randomly from each tissue section and positioned such that respective vessel lumina were in the centre of each field. Overall field-size dimensions were 190 j.m x 190 jim.
Colour detection thresholds were set for the red-brown colour of the DAB chromogen based on an intensely labelled point and a default colour threshold width was determined. The degree of labelling in each field was determined by the percentage of the field area minus acellular space within the default colour width as determined by the image analysis system. This was an approach similar to that used for canine tumours . For uniformity, comparison of labelling was restricted to those lumina whose diameters fell in the range 5-8 jim. Results from each tissue section were pooled to determine the means of area labelled in periportal and pericentral regions.
Binding of 2-nitroimWazoles to normal tissue GE Arteel et al 892 Results ELISA analysis Except for pimonidazole, 2-nitroimidazoles were weak competitors of solid phase antigens for pimonidazole antisera as measured by ELISA. Cross-reactivity data for CCI-103F are shown (Figure 1). Other 2-nitroimidazoles including misonidazole, benznidazole and etanidazole showed little or no cross-reactivity (data not shown). Enzyme-digested, pimonidazole-BSA adducts were approximately 25 times better than free pimonidazole as competitive inhibitors of antisera binding (Figure 1) but for convenience, pimonidazole was used as the standard. A correction factor of 25 was used to estimate the amount of pimonidazole bound to cell and tissue protein. Because of the relative insensitivity of the ELISA to free pimonidazole, analysis of tissue-bound pimonidazole could be carried out without interference from small amounts of free pimonidazole in tissue samples. Cellular proteins labelled with pimonidazole were digested into peptides before ELISA and it was assumed that adducts to small molecules such as glutathione were detected with the same sensitivity as protein adducts.
Oxygen dependence ofpimonidazole binding in EMT6 cells The oxygen dependence of pimonidazole binding to EMT6 cells in vitro showed half-maximal inhibition at a gas phase oxygen concentration of 1500 p.p.m. or 0.15% (Figure 2). Binding was measured by ELISA in terms of nmol pimonidazole bound per g of protein in the cell suspensions.
The ratio of binding under anoxia (12 p.p.m. P02) to that under air (2 x 105p.p.m. PO2) was 12:1 so that a wide dynamic range was available for the binding experiments. The concentration of oxygen that gave half-maximal inhibition of pimonidazole binding was similar to published values of 0.1-0.3% for the half-maximal inhibition of misonidazole binding . The protein content of the cell suspensions was typically 1.75 mg ml-' and the concentration of oxygen in the aqueous phase could be somewhat lower than that in the gas phase owing to oxygen consumption (Jones, 1985). The aqueous phase oxygen concentration was not measured, however, and the Km for inhibition of pimonidazole binding by oxygen must be considered to be approximate. While measurement of the aqueous phase concentration of oxygen would provide a more accurate value for the Ki,, the correspondence between pimonidazole and misonidazole binding was considered adequate for the purposes of the present experiment.
Intensity ofpimonidazole binding in liver tissue The overall intensity of pimonidazole binding to tissue from livers perfused at low flow rates when measured by ELISA of homogenised, protease-digested tissue was not significantly different in livers perfused in either anterograde (1.5 ± 0.8 nmol mg-' protein; n = 4) or retrograde (1.2 ± 0.3 nmol mg-' protein; n = 3) directions. In this analysis, no distinction was made between pimonidazole bound to small, acid-soluble molecules and pimonidazole bound to large, acid-insoluble molecules. It was found by UV analysis that digested homogenates contained free pimonidazole in the range of 23-79 JLM. The tissue-bound values were corrected accordingly.
Immunohistochemical patterns ofpimonidazole binding in liver As expected, immunostaining in livers perfused at low flow rates in the anterograde direction was localised in oxygenpoor, centrilobular regions ( Figure 3a). The pattern of immunostaining was reversed in liver perfused in the retrograde direction; that is, the staining was now localised in the oxygen-poor, periportal region of the tissue (Figure 3b). The only exception to this reversal of immunostaining when the oxygen gradient was reversed was the immunostaining of the single layer of cells directly adjacent to the pericentral vein following retrograde perfusion (Figure 3c). No such layer of immunostained cells was observed next to the periportal area following perfusion in the anterograde direction.
At high magnification (x 1000, Figure 3d), immunostaining increased from background levels to dense staining over 1-2 cell diameters. In lightly stained cells, the staining was distributed equally over the nucleus and cytoplasm. In more heavily stained cells, an intensified immunostaining was observed over cell nuclei.
Image analysis of immunostained liver sections Quantitative image analysis (Figure 4) revealed that the fraction of liver tissue immunostained in the periportal regions during perfusion in the retrograde direction was not significantly different from that observed for the pericentral regions during perfusion in the anterograde direction. The single layer of immunostained cells around the central vein did not contribute significantly to the fraction of tissue area labelled during perfusion in the retrograde direction.

Discussion
The microregional patterns of 2-nitroimidazole binding (Figure 3) are completely consistent with the distribution of oxygen in liver perfused at low flow rates. There is no evidence that specialised nitroreductase activity located in the pericentral region of liver dominates the binding process as Cobb et al. (1990a,b) suggested. The results presented here indicate that oxygen-dependent nitroreductase activity is homogenously distributed in liver so that 2-nitroimidazole binding will occur whenever regional oxygen concentrations decline to such a level that electrons flow to the nitroaromatic compounds rather than to oxygen. This conclusion is in agreement with the known distribution of reducing equivalents and nitroreductases in liver.
With respect to the distribution of reducing equivalents, previous studies have shown that the potential for oxygen uptake is similar in periportal and pericentral regions of liver (Matsumura et al., 1986). That is, the high rate of oxygen uptake in periportal regions during perfusion in the normal or anterograde direction is shifted to the pericentral region during perfusion in the retrograde direction. Importantly, about 85% of oxygen consumption in liver tissue is due to oxygen reduction by electrons from the electron transport chain (Matsumura et al., 1986). These are the same electrons that reduce nitroaromatic compounds in the absence of oxygen so it is not surprising that the potential for nitroimidazole reduction and binding (Figure 3), like oxygen uptake, is similar in pericentral and periportal regions of liver.
With respect to the distribution of nitroreductases, it is true that cytochrome P450-dependent nitroreductases are localised predominantly in pericentral regions (Jungerman and Katz, 1989), but other nitroreductases such as aldehyde dehydrogenase (Wolpert et al., 1973) are distributed throughout liver tissue (Kashiwagi et al., 1983). Enzyme distributions are not altered during liver perfusion experiments (Thurman and Kauffman, 1985) and it is unlikely that measurable enzyme induction would occur during 50 min of low-flow perfusion. Incubation under hypoxia for prolonged periods (> 8 h) is generally required to induce redox enzymes such as DT-diaphorase (e.g. Phillips et al., 1994) and subsequent, prolonged incubation under aerobic conditions is often needed for the changes induced under hypoxia to be detectable (e.g. O'Dwyer et al., 1994). This appears to be the case for liver tissue in vivo as well. For example, 8-9 days of chronic in vivo hypoxia induced by inhalation of 10% oxygen produced no detectable changes in rat hepatic cytochrome P450 content (Aw et al., 1991).
If regional cytochrome P450-dependent nitroreductase activity accounted for hypoxia marker binding in liver tissue perfused in the anterograde direction, then perfusion in the retrograde direction should leave the binding pattern unchanged. That is, pimonidazole bioreduction would be analogous to mono-oxygenation in phenobarbital-treated rats which follows the distribution of cytochrome P450s irrespective of the direction of perfusion (Thurman and Kauffman, 1985). However, the pattern of pimonidazole binding follows the distribution of oxygen in the tissue being completely reversed after perfusion in the retrograde direction (Figure 3a and b). Bioreductive activation of pimonidazole appears to Binding of 2-nitroimidazoles to normal etssue GE Arteel et al 894 fall, therefore, into the class of processes such as oxygen uptake for which underlying metabolic systems are available uniformly throughout the liver. In support of this conclusion, the extent of binding around the portal triad during retrograde perfusion is quantitatively similar to that around the central vein during anterograde perfusion ( Figure 4) and overall binding intensities as measured by ELISA are not significantly different for perfusion in either anterograde or retrograde directions. It is clear that low oxygen concentration rather than unique nitroreductase activity determines the distribution of 2-nitroimidazole binding in liver tissue.
Pimonidazole binding in the rim of cells around the central vein during perfusion in the retrograde direction is a minor but interesting exception (Figure 3c). The single layer of cells around the central vein is unique in other ways. For example, these cells are the only cells in liver that possess the enzyme glutamine synthetase (Gebhardt et al., 1988). This enzyme is involved in ammonia detoxification and pH homeostasis (Haussinger et al., 1986) but it would not appear to be capable of reductively activating 2-nitroimidazole compounds. It is possible that these cells possess unique nitroreductase activity in addition to specialised ammonia metabolism although further work will be needed to confirm this. In any case, the possibility that these cells or the microregional distribution of specialised nitroreductase activity accounts for the predominant pattern of hypoxia marker binding in liver is not supported by the present experiments.
It is known that intracellular uptake of weak bases such as pimonidazole (pK, 8.6 at 37°C) can be influenced by changes in extracellular pH. While normoxic liver has mechanisms for pH homeostasis (Haussinger et al., 1986), acidosis can accompany hypoxia/ischaemia in other tissues and it is appropriate to consider the effect that changes in pH might have on pimonidazole binding. For an extracellular pH of 7.3, pimonidazole concentration exceeds that in the surrounding medium by a factor of 2-3 for both rodent and human tumour cells (Dennis et al., 1985;Watts et al., 1990) but as the extracellular pH declines to 6.6, intracellular pimonidazole concentration falls to match that in the surrounding medium (Dennis et al., 1985). This decrease in intracellular concentration is accompanied by a parallel decrease in the intensity of pimonidazole binding to cellular glutathione and macromolecules. For example, a change in extracellular pH from 7.3 to 6.8 leads to a decrease in the concentration of glutathione and macromolecular adducts by a factor of 2-3 (JM Yates et al., 1995, unpublished). A decrease in extracellular pH in hypoxic regions could, therefore, decrease pimonidazole binding. However, this effect is small compared with the 12-fold difference in binding intensity between anoxic and aerated cells ( Figure 2) and, with respect to the present experiments, in the wrong direction since binding was observed to increase, not decrease, in the hypoxic regions of perfused liver. In the image analysis of hypoxia marker binding (Figure 4) it should be noted that, once the threshold intensity distinguishing stained from unstained cells is established, small differences in the intensity of pimonidazole binding among cells in the immunostained regions are not registered. The ELISA data do reflect intensity differences but, in fact, they revealed no significant difference in binding intensity between pericentral and periportal regions. Therefore, we conclude that changes in pH did not play a role in pimonidazole binding patterns observed in the perfused liver experiments. Hypoxia markers do not provide a precise measure of oxygen concentration in tissues but it is reasonable to assume, based on comparison with misonidazole, that halfmaximal inhibition of pimonidazole binding in vivo occurs in the range of 1-6 tM oxygen (Franko and Koch, 1984).
While this concentration range is well below the values measured in normal tissues with oxygen electrodes, the present results show that oxygen gradients are created on a cellular scale (Figure 3d) so that average tissue P02 is not a good predictor of 2-nitroimidazole binding under normoxic conditions. Hypoxia markers were developed primarily for use in tumours where the patterns of binding have been found to be consistent with the distributions of oxygen expected on the basis of the Thomlinson and Gray (1955) analysis of carcinomas of the human bronchus. Nevertheless, the hypoxia marker technique is dependent on nitroreductase activity in tissues and there is little control of this aspect of the assayparticularly in a clinical setting. The challenge posed by the observation that 2-nitroimidazoles bind to some normoxic tissues cannot, therefore, be ignored. It raises the mechanistic question of whether binding in all cases is due to low oxygen concentration and the practical question of whether normal tissue hypoxia will interfere with the use of hypoxia markers in tumours. From a mechanistic point of view, the demonstration that the binding of 2-nitroimidazoles in liver tissue is primarily dependent on oxygen concentration provides support for the idea that hypoxia markers, in general, reflect patterns of oxygen concentrations in normal and tumour tissue. From a practical point of view, this opens up the possibility that hypoxia markers will be useful in studies of liver pathophysiology associated with changes in liver oxygenation as appears to be the case for alcohol-induced liver damage Arteel et al., 1995). Hypoxia markers might also be used in identifying normal tissues at risk with respect to hypoxia-dependent cytotoxins or to radiation sensitisation by procedures designed to increase tissue P02* Carbogen breathing, for example, increases radiation sensitivity in skin (Rojas et al., 1992) as hypoxia marker binding would predict. Finally, the results of the present study support the premise underlying the development of the immunohistochemical, hypoxia marker method.
Non-invasive hypoxia marker techniques might be useful in following changes in tumour hypoxia, but a biopsy-based, histological investigation of tumour hypoxia is the only way to discriminate between hypoxia in tumour and surrounding normal tissue (Raleigh et al., 1987).