Quantification of longitudinal tissue pO2 gradients in window chamber tumours: impact on tumour hypoxia

We previously reported that the arteriolar input in window chamber tumours is limited in number and is constrained to enter the tumour from one surface, and that the pO2 of tumour arterioles is lower than in comparable arterioles of normal tissues. On average, the vascular pO2 in vessels of the upper surface of these tumours is lower than the pO2 of vessels on the fascial side, suggesting that there may be steep vascular longitudinal gradients (defined as the decline in vascular pO2 along the afferent path of blood flow) that contribute to vascular hypoxia on the upper surface of the tumours. However, we have not previously measured tissue pO2 on both surfaces of these chambers in the same tumour. In this report, we investigated the hypothesis that the anatomical constraint of arteriolar supply from one side of the tumour results in longitudinal gradients in pO2 sufficient in magnitude to create vascular hypoxia in tumours grown in dorsal flap window chambers. Fischer-344 rats had dorsal flap window chambers implanted in the skin fold with simultaneous transplantation of the R3230AC tumour. Tumours were studied at 9–11 days after transplantation, at a diameter of 3–4 mm; the tissue thickness was 200 μm. For magnetic resonance microscopic imaging, gadolinium DTPA bovine serum albumin (BSA-DTPA-Gd) complex was injected i.v., followed by fixation in 10% formalin and removal from the animal. The sample was imaged at 9.4 T, yielding voxel sizes of 40 μm. Intravital microscopy was used to visualize the position and number of arterioles entering window chamber tumour preparations. Phosphorescence life time imaging (PLI) was used to measure vascular pO2. Blue and green light excitations of the upper and lower surfaces of window chambers were made (penetration depth of light ~50 vs >200 μm respectively). Arteriolar input into window chamber tumours was limited to 1 or 2 vessels, and appeared to be constrained to the fascial surface upon which the tumour grows. PLI of the tumour surface indicated greater hypoxia with blue compared with green light excitation (P < 0.03 for 10th and 25th percentiles and for per cent pixels < 10 mmHg). In contrast, illumination of the fascial surface with blue light indicated less hypoxia compared with illumination of the tumour surface (P < 0.05 for 10th and 25th percentiles and for per cent pixels < 10 mmHg). There was no significant difference in pO2 distributions for blue and green light excitation from the fascial surface nor for green light excitation when viewed from either surface. The PLI data demonstrates that the upper surface of the tumour is more hypoxic because blue light excitation yields lower pO2 values than green light excitation. This is further verified in the subset of chambers in which blue light excitation of the fascial surface showed higher pO2 distributions compared with the tumour surface. These results suggest that there are steep longitudinal gradients in vascular pO2 in this tumour model that are created by the limited number and orientation of the arterioles. This contributes to tumour hypoxia. Arteriolar supply is often limited in other tumours as well, suggesting that this may represent another cause for tumour hypoxia. This report is the first direct demonstration that longitudinal oxygen gradients actually lead to hypoxia in tumours. © 1999 Cancer Research Campaign

mammary adenocarcinomas were transplanted onto a fascial plain of subcutaneous tissue at the time of window chamber surgery. After surgery, but before experimentation, the animals were housed individually in an environmental chamber maintained at 34°C and 50% humidity with continuous access to food and water. Imaging studies were performed 9-11 days after transplantation. The maximum tissue thickness was 200 µm. The surgical preparation and housing of chamber-bearing rats was performed at DUMC (Duke University Medical Center). Magnetic resonance imaging and intravital microscopy procedures were performed at DUMC. PLI studies were performed at the University of Pennsylvania. After implantation of the window chambers and tumour growth, the animals were shipped to the University of Pennsylvania within 2 days of PLI studies. All protocols were approved by the DUMC Institutional Animal Care and Use Committee.

Anaesthesia
Animals were anaesthetized with sodium pentobarbital (50 mg kg -1 i.p.) for all surgical and experimental procedures. Body temperature was maintained using Deltaphase Isothermal pads (Model 39 DP, Braintree Scientific, Braintree, MA, USA). Blood pressures were monitored using a digital manometer (FiberOptic Sensor Technologies, Ann Arbor, MI, USA) and averaged 90-100 mmHg, which is typical for this anaesthetic regimen in our hands (Dewhirst et al, 1996a;Shan et al, 1997) (data not shown).

Intravital microscopy
In a separate group of animals, visualization of the vasculature of both tumour surfaces was performed using intravital microscopy (Zeiss MPS Intravital Microscopy System, Zeiss, Thornwood, NY, USA). The window chamber tissue was imaged at 25× and images were recorded using a colour 3CCD video camera (model ZVS 3C75DE, Optronics Engineering, Goleta, CA, USA) and frames were captured with a frame grabber (model LG3PCIRS170, Scion Corporation, Frederick, MD, USA) installed on a microcomputer (model G6-233, Gateway 2000, Kansas City, MO, USA). These images were subsequently analysed using image analysis software (NIH Image).

Magnetic resonance microscopy
In a separate group of animals, three-dimensional visualization of tumour vasculature was obtained using magnetic resonance microscopy, with Gd-albumin administered as a contrast agent (Smith et al, 1994). Briefly, the upper glass window of the tumour chamber was removed from the anaesthetized rat (40-50 mg kg -1 , i.p. nembutal) and the skin flap was covered with phosphatebuffered saline (PBS). One millilitre of bovine serum albumin diethylenetriaminepentaacetic anhydride-gadolinium (BSA-DTPA-Gd) with approximately 1 mM Gd was injected i.v. Fifteen seconds after administration of the contrast agent, the skin flap was immersed in 10% formalin and surgically removed from the animal. The tumour sample was embedded in 3% agarose and placed in a 1-cm solenoid imaging coil. Imaging was performed on a 9.4-T Bruker Instruments magnet with an Omega System console (Freemont, CA, USA). Data were acquired using threedimensional spin warp encoding adapted for imaging large arrays. The array size by pixels was 256 3 , with pixels of 40 µm on a side [repetition time (TR) = 200 ms, echo time (TE) = 6 ms], and four excitations for each phase-encoding step.

Phosphorescence lifetime imaging (PLI)
PLI was used to measure vascular pO 2 , after i.v. administration of 3.5 mg Pd-mesotetra-(4-carboxyphenyl) porphyrin (oxyphore). Blue (419 nm) and green light (525 nm) excitations (BLE and GLE respectively) from the upper and lower tumour surfaces were made. The porphyrin has absorption peaks at both of these wavelengths, but the emitted phosphorescence spectrum and lifetime are independent of the wavelength of excitation. Phosphorescence was imaged using a grated, intensified CCD camera. The camera was turned on at varying times after the flash of excitation light. The rationale for using both wavelengths is that the penetration depth for blue light (approximately 50 µm) will be less than that for green light (>200 µm) because of differences in absorption by light absorbing chromophores, such as cytochromes haemoglobin and myoglobin. Details of these methods have been published previously (Wilson et al, 1992;Vinogradov et al, 1996;Cerniglia et al, 1997).
The animals were anaesthetized, oxyphore was administered i.v. via the tail vein and imaging commenced within 5-10 min of injection. The total imaging period lasted no more than 30 min, after which time the animal was killed using an overdose of nembutal. The size of the oxyphore complex is large and most of the signal observed is intravascular during this period of observation (Wilson et al, 1992). In the first ten experiments, a direct comparison between green and blue light excitation was made on the tumour surface of the chamber only. After it was seen that there were substantial differences for this comparison, an additional six animals were studied in which blue and green light excitation of both the tumour and fascial surfaces was made to further characterize the nature of the gradients observed.

Statistical methods
Differences in oxygenation parameters obtained using different excitation wavelengths (BLE vs GLE) or obtained from different surfaces of the window preparation (tumor vs fascial) were compared using the paired two-tail Student's t-test. Differences were considered significant when P < 0.05; all data are reported as means; standard errors of the mean (s.e.m.) are indicated in parentheses.

Orientation of tumour arterioles
The vascular structure of these tumours is quite different when visualizing it from either the fascial surface or the tumour surface ( Figure 1). On the tumour surface, the vasculature is composed of predominantly tortuous venular structures of low overall vascular density. In contrast, the fascial surface has a luxuriant growth of vasculature with arterioles that are readily visible. Characteristics of arterioles that distinguished them from other microvessels were: (1) relatively straight course, with divergent flow; (2) visible evidence for muscular wall; and (3) divergent branch angles that are near 90°. These arterioles can be observed to traverse along the fascial surface beneath the tumour. Typically, one or two of these vessels will enter the tumour parenchyma and at the point of entry visual evidence of smooth muscle was frequently lost. The magnetic resonance microscopic images substantiated the orientation of tumour-feeding arterioles (Figure 2). The vasculature of the fascial layer was quite dense. Large venules and/or arterioles with diameters greater than 40 µm (pixel size of images) were readily observed in these vascular networks. The upper surface of the tumour also showed a dense network of vessels, but the diameters were primarily less than or equal to 40 µm. In cross section, it was readily apparent that the vascular density in the interior of the tumour was less than that on either surface. However, microvessels with diameters >40 µm could be visualized penetrating through this space. Three tumours were studied in this way, but only one is shown for illustrative purposes because the others looked quite similar.

Phosphorescence lifetime imaging
Visualization of tissue pO 2 distribution can be set so that the entire range of values can be seen in the same image. When this has been carried out, gradients in pO 2 , from the normal tissue periphery to the interior of the tumour, are easily visualized (Figure 3). With blue light excitation (BLE), the nadir of these gradients was greater in relative size when visualized from the tumour surface as opposed to the fascial surface. In the example shown, the tumour surface appears to be more hypoxic with BLE than GLE. In contrast, BLE of the fascial surface shows less hypoxia than when the same preparation is illuminated with green light. If the colour range for pO 2 is restricted to emphasize the range from the periphery to the centre of the tumour, then intratumoral heterogeneity in pO 2 distribution is more easily seen (Figure 4). The average median pO 2 of the 16 tumours that were imaged from the tumour surface was 25 (±4) mmHg when measured with BLE and 32 (±4) mmHg when measured using GLE (Table 1). This difference was not statistically significant. However, there were significant differences when various parameters indicative of the lower end of the pO 2 distribution were compared. There were statistically significant differences in the 10th and 25th percentiles and the per cent of pixels <10 mmHg when comparing BLE and GLE (P < 0.03) with BLE, indicating a greater level of hypoxia than GLE. For example, the average percent of pixels < 10 mmHg was 22.1 (±5.6) for BLE vs 9.7 (±3.2) for GLE.
The average median pO 2 of the fascial surface was 40 (±7) mmHg when measured with BLE and 26 (±4) mmHg when measured with GLE. For all of the parameters studied, there was a trend towards a greater level of oxygenation with BLE than with GLE, but none of the comparisons were statistically significant. In the six window chambers where both surfaces were imaged, the gradient was observed more clearly. For example, with BLE the percent pixels <10 mmHg averaged 36.3 (±8.9) from the tumour surface vs. 6.9 (±4.0) when imaged from the fascial surface (P < 0.05). Similar differences were seen for the 10th and 25th percentiles of the frequency distribution. In contrast, the more deeply penetrating GLE did not show statistically significant differences when comparing tumour and fascial surfaces in the same six preparations.

DISCUSSION
These data clearly show that there is a longitudinal gradient in tissue oxygenation in the window chamber tumour model that is set up by the geometric orientation of arterioles entering the tumour from one surface. The existence of the gradient was demonstrated because of differences in the depth of penetration of blue and green light into the window chamber tissue. When BLE of the tumour surface was used, the tissue appeared more hypoxic than with GLE because a thinner region of the tissue at the upper surface was being sampled with BLE. GLE penetrates completely through the tissue, thus providing signals from the whole thickness of the window. Similarly, in the subset of tumours in which both surfaces were illuminated with BLE, the tumour surface consistently appeared more hypoxic than the fascial surface.
It has been suspected for some time that the relative lack of arteriolar input into tumours may be a contributing cause to tumour hypoxia, but this study is the first to prove that limited arteriolar supply contributes to longitudinal gradients that result in tumour hypoxia. The relative lack of arteriolar supply to tumours has been discussed by a number of authors. For example, Lindgren and Docent (1945) performed extensive histological analysis of benign and malignant human tumours and found relatively little evidence for vascular smooth muscle in tumour vessels of either classification. Similar observations were made by Lagergran et al (1960) in examining a series of human sarcomas. These authors also performed angiography and noted the relative lack of arteriolar input to the tumours (Lagergren et al, 1960). Many other similar studies have been made in murine tumours, as reviewed by Warren (1979). Perhaps the most detailed analysis of afferent blood supply of murine tumours was made by Falk (1978Falk ( , 1980. The threedimensional orientation of vessels in tumours was recreated using a benzidine staining procedure of serial histological sections. With this method, he was able to trace the path of arterioles from the surrounding normal tissues and described the same type of afferent patterns that we describe herein. In many normal tissues, there is a redundancy in arteriolar supply, or at least a repeated unit design in which the tissue volume supplied by a single arteriole is tightly regulated (e.g. the kidney). The relative number of arteriolar vessels is high enough that some authors have suggested that a major source for oxygen transport in some tissues is the arterioles, rather than the capillaries (Ellsworth and Pittman, 1990;Kerger et al, 1995;Torres et al, 1996). Longitudinal gradients also exist in normal tissues, but, because of the redundancy of vascular supply, such gradients do not lead to vascular or tissue hypoxia in most cases (Duling and Berne, 1970;Ellsworth et al, 1987;Swain and Pittman, 1989). The evidence from this paper and many others suggests that the volume of tissue supplied by a single artery in tumours can be much greater than for normal tissues. The anatomical constraints of a limited arteriolar supply being constrained to the periphery of the tumour is what allows for the steep longitudinal gradient leading to vascular and tissue hypoxia, as we have described in this paper.
The most frequently cited reason for chronic hypoxia in tumours is the presence of long intervascular distances (Thomlinson and Gray, 1955). We began to suspect that the origins of chronic hypoxia were more complex when we measured intravascular pO 2 in tumour vessels of the tumour surface of window chamber tumours, using recessed tip microelectrodes. Those results showed that these microvessels had perivascular pO 2 values averaging around 10 mmHg near the centre of the tumour mass (Dewhirst et al, 1992a). Twenty-five per cent of such vessels had pO 2 values below the level of detection (1 mmHg), in spite of the fact that they had blood flowing through them. Since that time, others have found corroborating evidence for the existence of hypoxia in functional tumour vessels. For example, Helmlinger et al (1997) recently demonstrated vascular hypoxia in window chamber tumours using the phosphorescence-quenching method. Fenton and Boyce (1994)  Phosphorescence lifetime imaging of the tumour surface with thresholding to compare the entire pO 2 distribution (left column) vs. restriction to the range within the tumour (right column). Spatial heterogeneities in tumour surface pO 2 are more easily seen when the range is limited to pO 2 values between 0 and 40 mmHg as haemoglobin saturation measurements indicative of hypoxia (Fenton and Boyce, 1994). The authors suggested that this may be due to transient vascular stasis between the time that the dye was given and tissue removal for cryospectrophotometry. However, an equally plausible explanation is that the vessels were functionally hypoxic, even though they had active flow and contained red cells.
However, as yet, there has been no conclusive proof that vessels with blood flow are hypoxic in spontaneous tumours. Immunohistochemical studies of patterns of tumour hypoxia, using hypoxia marker drugs, in canine and human tumours have typically reported that most intense binding occurs near regions of necrosis, which are typically farthest removed from the nearest microvessel (Cline et al, 1990;Kennedy et al, 1997). Immunohistochemical localization of microvessels would help to clarify this issue further because it is frequently difficult to positively identify microvessels on the basis of H and E staining alone. Alternatively, cryospectrophotometry studies in human tumours have demonstrated ample evidence for vascular hypoxia, but it was not known if such vessels had active blood flow at the time the tissue was removed (Mueller-Klieser et al, 1981).
Finally, it is of interest to compare the pO 2 values obtained in this study with our prior microelectrode studies with this same tumour model. We previously found that the overall average pO 2 of tumour vessels was near 20 mmHg (Dewhirst et al, 1992a), similar to what was observed in this study using the phosphorescence quench imaging method with BLE. In addition, the average pO 2 of tumour arterioles was over 32 mmHg (Dewhirst et al, 1996a), which is similar to the average pO 2 of the fascial surface of these preparations in the current study using BLE. Finally, we observed a pO 2 gradient when comparing the normal tissue surrounding the tumour to its periphery and centre. Similar results were shown in the current study. The fact that these two independent methods of oxygenation measurement provide similar results makes the case stronger that the magnitude of the measurements reflects the true pathophysiological state. It is becoming increasingly clear that chronic hypoxia comes from several sources. Although large intervascular distances play a role, as was described by Thomlinson and Gray (1955), there are other features of tumour microvascular function that may contribute. These include arteriolar pO 2 deoxygenation (Dewhirst et al, 1996a), plasma channels (Dewhirst et al, 1996b), rheological effects (Dewhirst et al, 1992b), irregularities in tumour vascular geometry (Secomb et al, 1993) and oxygen consumption rates that are out of balance with the ability of the vasculature to deliver oxygen (Secomb et al, 1995). In this paper, we have added another feature to this complex set of circumstances that may contribute to chronic hypoxia in tumours. The features of tumour angiogenesis that are not permissive for arteriolar ingrowth into the tumour create steep longitudinal oxygen gradients, which contribute to intravascular and tissue hypoxia.