Variable presence of hypoxia in M006 human glioma spheroids and in spheroids and xenografts of clonally derived sublines.

Recently we reported the variable presence of hypoxia adjacent to necrosis in human glioma lines grown as subcutaneous tumours in severe combined immunodeficient (SCID) mice. To assess the basis for this observation, we examined the pattern of oxygenation in M006 and M006XLo glioma spheroids. We found a wide range of binding of [3H]misonidazole to cells adjacent to the necrotic core, analogous to the patterns seen in xenografts, indicating substantial differences in the central oxygen tension of the spheroids. Clonal selection was used to isolate single cell-derived sublines of the M006XLo line. Some sublines gave spheroids that showed narrow distributions of [3H]misonidazole binding to the cells adjacent to necrosis, whereas other sublines showed a range of binding similar to that seen in spheroids of the parent line. After additional passages in monolayer culture, clonal sublines occasionally gave rise to spheroids in which the mean oxygen tension of cells adjacent to necrosis differed substantially from that of the initial spheroids. No relationship was evident between the thickness of the rim of viable cells and the presence or absence of central hypoxia, over a wide range of rim thickness. These results indicate that different oxygenation characteristics of glioma spheroids and tumour microregions are unlikely to arise from stable genetic variants coexisting in the parent line.

The existence of hypoxic cells in malignant gliormas is suggested bv the extreme radioresistance of these tumours. by the presence of extensive areas of necrosis in grade IV tumours and by direct measurements of oxvyen tension on anaesthetized patients using microelectrodes (Rampling et al. 1994). However. no evidence of hN-poxia A-as detected in a series of 11 brain tumours usinc a bioreductivelx activated marker for hxpoxia. [I-I]iodoazomycin arabinoside (Urtasun et al. 1996). To assess potential explanations for this discrepancy. we examined the microregional distribution of a similar hypoxia marker ['H]misonidazole. in subcutaneous xenografts of several human glioma cell lines (Parliament et al. 1997). Ex-idence for severe hypoxia adjacent to necrosis A-as found consistentlx in MOlOb tumours. similar to that seen in rodent tumours (Chapman et al. 1981: Franko et al. 1992) and in sexveral types of human tumours labelled in situ (Urtasun et al. 19861. Hoxwever. in N1006 and M059K tumours approximately half of the necrotic areas w-ere not associated wxith hyIpoxia. This obserxation prox-ides support for the possibility that rumour cells adjacent to many regions of necrosis in human brain tumours are also A-ell oxy genated. We proposed an interpretation of the variable presence of hypoxia in some xenografted glioma lines (Parliament et al. 1997). A hich inxolx ed x-ariable rates of oxy gen consumption based on the concept of tissues responding to reduced oxy gen lexels as either Hochachka et al. 1996). Regions of tissue that possess a conxentional oxy gen-regulating phenotype would consume oxygen at a characteristic rate until they became sexerely hypoxic. but cell death would occur only at a greater distance from capillaries w here Glucose xxas depleted to a critical level. In contrast. regions of tumour tissue that exhibit an oxy gen-conforming phenotype would respond to moderately low oxygen lexvels (which are far from lirmiting metabolically) by reducino their oxygen consumption to low lex els: consequently. wxell-oxy,genated cells beyond the diffusion distance of glucose xould die. One of the hallmarks of oxx en-conformine, tissues is that they may reduce their rate of oxygen consumption without increasing glycoly sis. that is the Pasteur effect may be absent or minimal (Hochachka et al. 1996). An alternatixe interpretation of the xenograft data that does not require a reduction in oxygen consumption is that those regions of tissue that contain well-oxvaenated necrosis exhibit an enhanced Pasteur effect at moderately low-oxy gen levels. x-hich leads to a much shorter diffusion distance for Glucose than for oxvgen. This xould require that well-oxy genated cells exentually die wxhen deprived of Glucose (Hlatky et al. 1988). Another factor that might contribute to the obserx ations is the possible egress of toxic metabolites from xithin necrotic regions (Freyver. 1988).
Recently. using monolayver cultures wxe hax-e found exidence that supports our oxy gen regulator/conformer hypothesis (Allalunis-Turner et al. 1998). Cells of the M059K and M006 lines that were exposed to 2%7 or 0.6% oxy gen for 4 days exhibited markedly reduced oxvaen consumption. consistent with our hypothesis that they hax-e the potential to act as oxy gen conformers. Howxever. cells of the MOl Ob line behaxed as oxy gLen regulators. with similar rates of oxvyen consumption in air. 2% and 0.6%c oxygen. as predicted by our hypothesis.
The multicellular spheroid (Sutherland. 1988) is an excellent model system for assessing the foreaoing interpretations because it retains the three-dimensional structure of tumour tissue. vet prox ides a well-defined path for diffusion of nutrients and a geometrv that facilitates diffusion calculations. We grexx spheroids from the M006 cell line and found that the cells adjacent to the necrotic centre of the spheroids exhibited either substantial or minimal binding of ['H]misonidazole. which w as similar to the labelling of perinecrotic cells in M006 xenografted tumours. In this report we prox ide quantification of autoradiographic grain densities over the innermost spheroid cells. as well as a calibration curve of g,rain densitv as a function of oxv2en concentration. which permits estimates to be made of the oxy gen tension w ithin spheroids. The profiles of bound misonidazole were determined in some spheroids to proxvide a comparatixve assessment of the rate of oxyven consumption in spheroids w ith different degrees of hypoxia in the innermost cell layers. An important question is w hether the variable presence of hypoxia is the result of stable genetic variants that co-exist in the M006 line. or whether it constitutes evidence of the apparently random operation of a mechanism that regulates the metabolic pathways that underlie the differences in hypoxia. If the latter explanation were correct. it would suggest that the growth conditions encountered in xenografts and in spheroid culture allow the regulatory system to affect the cells in an entire tumour microregion or spheroid in a coordinated manner. To decide between these possibilities. we initiated sublines from single M006 cells. grewx spheroids from sexeral stages of passage of those sublines and analysed their patterns of binding of ['H]misonidazole.

Cell line
The M006 glioma cell line used in these studies was derived from portions of a diagnostic biopsy obtained from a patient with a urade IV astrocytoma. and xxas supplied by Dr RS Day III. The tumour exhibited a relatively low mitotic index. prominent xascular endothelial proliferation and extensixve areas of necrosis. Details of the techniques used to establish the cell line haxe been published (Allalunis-Tumer et al. 1991). The M006X subline was derixed from a xenografted tumour grown in a SCID mouse and disaggregated by mechanical and enzymatic methods. as described prexiously (Parliament et al. 1997). The cultures were maintained as monolayers in minimal essential medium (MEM) with 12% fetal calf serum (Gibco. Grand Island. NY).

Spheroid growth
Cells were detached from dishes using, trypsin (Gibco). and I0W cells were suspended in 75 ml of medium in a 250-ml spinner flask. which xxas spun at 150 r.p.m. The gas phase was 95% air/5%7c carbon dioxide. Sexeral hundred small acggregates of cells fonned spontaneously within a few days. The medium was replenished by remoxing and replacing 50 ml on a schedule that depended on the apparent total number of cells and the colour of the medium. The first feeding typicall) occurred 1 week after initiation. and subsequent feedings were at progressixely shorter interxvals. w-hich were reduced exentually to 24 h. The gas phase was replaced aExduding two spheroids with a mixture of high and low grains ( Figure 3B).
completely at each feeding. The number of spheroids w-as reduced at each feeding to stabilize the rate of consumption of nutrients. The first spheroids reached diameters of 0.8-1.2 mm. at x-hich they w-ere eliaible for experiments. w ithin 4-5 eeks.

M006XLo subline
In an effort to denrve a subline that w-as adapted to grow-th at lowambient oxy gen. the M006XLo subline x as established from small M006 spheroids that had been exposed continuously to 0.6%7 oxygen for 13 days. beainning 1 week after initiation in air. The spheroids were disaggregated wxith 0.25%c trypsin. and the cells w ere groxx n in monolayer culture for four passages (5 w eeks). then injected subcutaneously in SCID mice. A tumour x-as disaggregated to yield a new line designated M006XLo. w hich w as propagated in monolayer culture. Cells from the third passage were stored in liquid nitrogen. and each line x-as discarded after 3 months in culture and replaced from frozen stock.

Clonally derived sublines
Sublines of M006XLo xxere established as follows. To form a feeder layer. cells were plated at a density of 100 cells per A-ell in 24-well culture dishes (Coming Cell Wells. Coming Ltd. Corning. NY). The follow inr day the feeder lax er w-as irradiated to a dose of 12.5 Gy 11.4 Gv min-' ) using a ",-Cs irradiator (Shepherd Mark I. Shepherd and Associates. San Fernando. CF. USA). Subsequently x-arious dilutions of non-irradiated cells were plated to vield mean British Joumal of Cancer (1998) 78(10). 1261-1268 frequencies of cells per well between 0.25 and 2. No colonies w-ere obserxed in 24 control x-ells that contained onlv irradiated cells.
Dilutions that xielded more than one colony per four wells A-ere discarded. to minimize the probability of selecting a colonv that arose from more than one cell. To further reduce this probabilitv. all eligible w-ells w-ere examined microscopically for the presence of colonies at I and 3 w-eeks. and anv wells that appeared to contain multiple colonies w-ere not used. Eligible colonies at 3 weeks A-ere removed with trvpsin and sublines were established and designated by consecutix e numbers. The source for the sublines x as either the original M006XLo line in monolayer culture. or cells that were obtained directlv from M006XLo spheroids trypsinized at a diameter of 0.8-1.2 mm. In the latter case the letter 's' is added to the numerical designation (e.g. subline 5s).
The passage number of each subline when it was used to initiate spheroids is designated in parentheses follow-ing the subline designation. The number following the letter 'p' indicates the number of passages following trypsinization from the multixxell dish [e.g. subline 5s (p4)]. If the subline A-as frozen at that point. the letter 'f is added. and the number of passaaes following thawing is indicated by a subsequent number [e.g. 5s(p4fl fl.

Labelling of hypoxic cells ['H]Misonidazole
A-as synthesized following a published procedure (Born and Smith. 1983) and stored in ethanol. Spheroids w-ere labelled at a final concentration of 50 jiNi and a specific activity of between 400 and 900 jCi mgo-1. In order to maintain a stable nutritional environment for the spheroids that were to be labelled under the conditions of growth. 2 days before labelling 25 spheroids of the appropriate diameter were selected manually and incubated in a separate spinner flask that w-as gassed continuously w-ith 95% air/5%7 carbon dioxide. In most experiments as many as 100 additional spheroids were placed in a separate. gassed flask for labelling at reduced oxy en levels. The ethanol was evaporated from the required quantity of stock solution of radioactive drug. and the ['H]misonidazole w-as dissolved in medium removed from the flask containing, 25 spheroids. For labelling under growth conditions. 25 ml of the medium was returned to that flask (after removing the remaining non-radioactive medium). and gassing continued during incubation. Labelling under reduced oxy gen levels was performed in aliquots of the same medium in glass Petri dishes in sealed aluminum chambers on a shaker table. as described previously (Franko et al. 1987). The desired oxygen level was obtained by a series of partial evacuations. each of which was followed by refilling of the chambers with 95% nitroaenlr5% carbon dioxide. Metabolic actixation of misonidazole during the degrassing procedure w-as minimized by precooling the medium and the chambers to 0°C and maintainingr this temperature during, the 30to 40-min degassing process. The incubation period was 3 h for the spinner flask. and an additional 30 min for the aluminum chambers. w-hich is the time required for the medium in the dishes to return to 37' in the cabinet used for incubation. The oxygen tension in each chamber was measured at the end of the incubation period. as described previously (Franko et al. 1987).

Autoradiography
The spheroids were fixed in formalin and embedded in wax. and 4gm serial sections w-ere taken throughout the spheroids. Slides w-ith sections near the centres of the majority of the spheroids were A Results for individual spheroids labelled in air. B Means of data from spheroids Labelled in air, pooled as the two spheroids with low grain density adjacent to necrosis (circles) and the four spheroids with high grain density adjacent to necrosis (inverted triangles). Means of data from five spheroids labelled in nitrogen (squares) dipped in NTB-2 nuclear track emulsion (Kodak. Rochester. Newx York) diluted 1:1 with distilled water and stored with dessicant for 1-4 weeks. the emulsion was then developed and fixed and the tissue was stained with haematoxylin and eosin.
Quantification of radioactivity Grains were scored manually at an overall maanification of x 1000.
using a grid of squares that measured 10 gm per side. For most experiments grains were scored only over the outermost cell layer and over cells adjacent to necrosis. Locations for scoring were chosen systematically around the circumference of the spheroid or the necrotic centre. and a total of approximately 500 grains were counted per region per spheroid. The mean grain density (grains per 100 tm') was calculated at the two locations. The density of background grains. which was subtracted from the grain density over spheroid tissue. was determined individually for each slide in the vicimity of the spheroid sections scored. For most of the analyses the ratio of the grain densities at each location for each spheroid was the quantity' employed in estimating oxygen tension and in statistical analysis. In one experiment the distribution of British Joumal of Cancer (1998) 78(10) grains across the rim of intact cells was determined by scoring arains on radial tracks. Care was taken to ensure that the sections chosen passed near the centre of the spheroid. A total of 12 tracks (eight tracks for spheroids labelled in nitrogen) from several sections were scored for each spheroid. Because embedding and sectioning artefacts distort the spheroid. the thickness of the rim of intact cells may be expected to vary somewhat for different tracks on the same spheroid. To average only the data taken at essentially the same location in a spheroid using tracks of different lengths.
the mean length (in grid divisions) of all tracks was determined and divided by two. This number of grid divisions was averaged from the spheroid surface inwards. and the same number of grid divisions was averaged from the edge of necrosis outwards. At the mid-point. where the two sets of means were joined. some central grid squares were ignored on tracks longer than the mean. whereas in tracks shorter than the mean a few grid squares were counted twice.

Xenografted tumours
Tumours of selected sublines were grown subcutaneouslv in SCID mice. and at least four tumours of each subline were labeled with [ H]misonidazole. The tumours were fixed in formalin and the distribution of misonidazole binding was assessed usinr autoradiography as described above for spheroids. Procedures specific to the xenografts have been described recently (Parliament et al. 1997).

RESULTS
Spheroids derived from the M006 line uniformly displayed central necrosis at diameters of 800-1200 gm. and exhibited patterns of labelling with [H]misonidazole that were similar to those of xenografted tumours (Parliament et al. 1997). Within the same flask. in some spheroids the grain density above cells adjacent to necrosis was essentially the same as the grain density at the spheroid surface. whereas in other spheroids the grain density at the edge of necrosis was as much as tenfold greater than that at the spheroid surface. Spheroids from the M006XLo line gave similar results. Examples of the pattern of grain density as a function of distance from the spheroid surface are shown in Figure IA. The six most extreme patterns seen in a population of 20 M006XLo spheroids were quantified in detail along 12 radial tracks. Error bars have been omitted for clarity: however, the 95% confidence limits were typically smaller than 50% of the mean for grain densities less than 5 per 100 gm'. whereas for grain densities greater than 25 the 95% confidence limits were usually less than 25% of the mean. A sample of spheroids was labelled in nitrogen in most experiments. For the M006 line a 3-h exposure to anoxia caused extensive necrosis in the central region of the rim. based on comparison with spheroids labelled in air. The original central necrosis was clearly distinguishable. and the cells in the 4-6 layers adjacent to this necrotic region exhibited a mixture of appearances from normal to clearly necrotic. The cells in the outermost three layers appeared to be normal. whereas all cells between these two layers appeared to have died. In M006XLo spheroids exposed to anoxia only two regions were evident. The outermost 3-5 cell layers appeared to be unaffected. whereas at greater depths the frequency of isolated necrotic cells was noticeably increased. When grains were scored over spheroids labelled in nitrogen. obviously pyknotic or necrotic cells were avoided. The mean distribution of  Figure lB. The 95%7 confidence limits for the nitrogen data were typically less than 15% of the mean value. The same data for labelling in air are shown in both panels of Figure 1.
The relationship between misonidazole binding and oxygen tension for M0)6XLo spheroids is shown in Fioure 2. The mean grain density over the outermost layer of cells is plotted for 100 determinations (40 for nitrogen) on at least 15 different spheroids at each oxygen level. For each spheroid several different sections on two or three slides were used. The line was fitted by linear regression to the logarithms of the grain densities and oxygen tensions (excluding nitrogen). and the resulting equation (based for convenience on units of thousands of parts per million for oxygen tension) gave a slope of -0.7144 (95% confidence interval -0.92 to -0.50) and an intercept of 2.464 (95% CI 2.19-2.74). and a correlation coefficient (r) of 0.99 1.
An estimate of the oxygen tension in cells at the edge of the necrotic region can be obtained by inserting the grain density over these cells into the foregoing equation. However. many variables affect the grain density. in addition to the oxygen tension.
including the specific activity of the [jH]misonidazole. time of exposure and thickness of the emulsion. and humidity during exposure. As some of these variables are unknown. it is necessary to adjust (normalize) the vertical position on Figrure 2 of either the calibration curve or the experimental data by matching grain densities at locations where the oxygen tension is known. Two such locations are available. which give two independent estimates of the normalization factor. First. the mean grain density over the innermost cells of spheroids labelled in nitrogen may be compared directly with the nitrogen point in Figure 2. The ratio of those quantities constitutes one normalization factor. Second. the oxygen tension at the spheroid surface may be assumed to be equal to the value in the medium. Thus. the ratio of the grain density calculated from the equation for 20% oxygen (Figure 2) to the grain density observed at the surface of spheroids labelled in air gaives a second normalization factor. For cons enience. in this case the experimental data were normalized to the calibration curve. The observed grain density over cells adjacent to necrosis was multiplied by the ratio of the calculated grain density in 20% oxygen ( Figure 2) to the observed grain density at the spheroid surface. In effect. this calculation involves multiplying the grain The mean grain density over the innermost cells was normalized by division by the mean grain density over the outermost cell layer. Each point represents this ratio for one spheroid. Also shown is the oxygen tension corresponding to the given range of grain density ratios. See text for explanation of the passage designation in parentheses. densitx on the calibration curve at 20c% oxv2en bv the ratio of the arain densitx measured oxer the cells adjacent to necrosis to the ,ran density oxver the outermost cell laver. This ratio can be calculated individually for each spheroid. which facilitates statistical comparisons of oxygen tensions estimated for different spheroid populations. and eliminates sexeral sources of experimental error that diminish the accuracv of the normalization using the data from incubation of spheroids in nitrogen. Spheroids from 13 clonally derixed sublines of MOO6XLo were grow-n successfully and labelled with ['H]misonidazole. In most cases an early passage after clonal selection w as used for the first attempt to groxx spheroids. and samples of an earIx passage were stored in liquid nitrogen. Subsequent spheroid populations were grown from cells recoxered from frozen storage. The arain density was quantified for spheroids labelled in air and expressed as the ratio of the grain density oxer cells at the edge of necrosis to the grain density at the spheroid surface ( Figure 3). Before freezing.
four sublines showed a wide range of grain densitx similar to the range seen in spheroids of the parent M006 and M006XLo lines. This type of distribution is illustrated in Figure 3A for subline 9s p2). Also showxn is the oxygen tension calculated from the calibration curve. Seven sublines gave an appreciably smaller range of grain density oxver the innermost cells. and the data from four of these lines are shown in Fiaure 3A. Two sublines were used only after retriexal from frozen storage. Txxo of the sublines that initially gresspheroids w-ith narroudistributions of grain density exhibited dramatic changes upon retriexal from frozen storage and during subsequent passage in monolayer culture. as show-n in Fiaure 3B. Subline 2(p3) initially gaxe the narrow est distribution obser ed. with a grain densitv ratio indicatix e of sufficiently severe hypoxia to confer appreciable radioresistance. After thawing. passage p3ft2 gave 18 spheroids with a relativelv narrow distribution of grain density consistent with central oxygen tensions well aboxe 3%'c. whereas two spheroids clearly contained regions x ith high and low g,rain densities on opposite sides of the necrotic centre. for which the mean grain density ratios are plotted.
Twxo of the sublines with initial narrox distributions shoxed relatively small changes subsequently. The 12s subline changed little upon retrieval from frozen storage. and appeared to drift tow-ards lower central oxy gen tensions with further passage. as showxn in Figure 3C. Oxver a similar history in culture the 5s subline changed little in mean arain densitx or in the wxidth of the distribution of grain density ratios (data not shown). The 14s subline. which was not assessed before to freezing. gaxve nearly identical distributions of grain density ratios from txo xials of frozen cells that were thaxed at times separated by almost a y-ear ( Figure 3C). Table 1. Spheroid batches that gave extremely heteroceneous distributions of grain density (e.g. subline 9s (p). Figure 3A) are not included in Table 1. The most reliable estimates werejudaed to be derived from the gorain density ratio betxxeen the innermost cells and the outer cell laver. and 95% confidence limits for these estimates are shown. These limits do not include the uncertainty in the equation for the calibration curve (Figure 2). and thus are useful principally for comparinc different experiments.

Estimated mean oxygen levels of the innermost cells of spheroids are shown in
The absolute values of the oxyven tensions are substantiallv less certain than the confidence limits indicate. When available. estimates are also riven that wxere derived using the data from the spheroids incubated in nitrogen to normalize the grain density over  Figure 3) plotted as a function of the thidess of the nm of morphologcally intact cells on the periphery of the spheroid. A Spheroids of subline 9s (p2). B Open circles, subline 13s (p3f1); solid circles, subline 13s (p2); open squares, subline 13s (p3f7); solid diamonds, subfine 2 (p3); open diamonds, subline 2 (p3f2) experimental error, and do not lend themselves to statistical analysis. Whereas substantial discrepancies are evident in the absolute oxygen tensions estimated by the two techniques. there is good agreement between the two independent estimates of the central oxygen tension regarding the direction, and moderate agreement regarding the magnitude of differences in oxygen tension among different passages of the same subline.
The data in Figure I suggest that in spheroids of the parent M006XLo line those spheroids with the highest grain density tend to have slightly thicker rims of intact cells than do spheroids with the lowest grain density. Measurements of the rim thickness plotted as a function of grain density are shown in Figure 4A for spheroids from the 9s(p2) subline. which gave a wide range of grain density (Figure 3a). Whereas some of the spheroids with low grain densities had rims that were thinner than those of the group of spheroids with high grain densities. the two distributions overlap substantially. Rim thickness measurements were made on all spheroids. and the largest differences were found in the two sublines that exhibited the largest variations in oxygen tension. as shown in Figure 4B. It is evident that in these sublines there is no relationship between the thickness of the rim of morphologically intact cells and the degree of hypoxia in the cells adjacent to necrosis. nor was any relationship apparent in the other sublines (data not shown).
Tumours were grown in SCID mice from the following sublines. with the passage number of the cells injected in parentheses: 2 (p3f2): 5 (p4fl): 13s (p3f2): 12s (p3f2): 5s (p4f2): and 14s (p2f2). The range of grain density over cells adjacent to necrosis after labelling with [IH]misonidazole was similar in all tumours and was indistinguishable from tumours of the parent M006XLo line. Thus. although in several cases spheroids that were initiated from the same passage showed a restricted range of grain density over the innermost cells (Figure 3 and Table 1). growth of those cells as tumours yielded the full range of phenotypes.

DISCUSSION
The question of the presence of severe hypoxia within human gliomas remains controversial. Our recent observations on the patterns of binding of hypoxia markers to human glioblastoma xenografts indicate that the oxygen tension at the boundary between necrotic tissue and viable cells can vary substantially (Parliament et al. 1997). As the mechanism by which this diversity arises is unknown. the relevance of this observation to the conditions within glioblastomas in human brain remains to be determined. The present work was undertaken to address the question of whether the variations in hypoxia adjacent to necrosis could be ascnrbed to the existence of stable genetic variants within the glioma lines.
The principal objective of this work was to compare the oxygen tension in different glioma spheroids. as indicated by the quantity of [ H]misonidazole bound to cells adjacent to the necrotic centre.
Although the data are plotted in terms of both grain density and estimated oxygen tension. it is important to note that the absolute level of the oxygen tension estimates is subject to substantial uncertainty. The calibration curve (Figure 2) was a single determination performed for comparison with two curves obtained in the same manner using fragments of glioma xenografts (Parliament et al. 1997). Although there is good agreement among the three data sets. which enhances confidence in the reliability of the data. the error limits on the regression line are substantial. The principal method for normalizing the data from each experiment to the conditions used to obtain the calibration curve was based on the ratio of the grain density at the surface of the spheroid to the grain density over cells adjacent to necrosis. Thus. for each spheroid the experimental data and the data for normalization were collected on portions of the same emulsion separated by a distance of less than 0.4 mm. This minimizes the potential sources of systematic error inherent in the autoradiography technique. However. because the calibration curve did not include a point for air. it was necessary to assume an oxygen concentration for the outermost cell layer. Although 20% oxygen was chosen for simplicity. it is known that the oxygen concentration at the spheroid surface is somewhat lower (by 3-7%) as a consequence of the unstirred layer of medium at the spheroid surface (Mueller-Klieser and Sutherland. 1982). If the correct value were known. its use would lower all of the estimated oxygen tensions slightly. The other normalization procedure compared the nitrogen point on the calibration curve with the grain density over the innermost cells of the spheroids Bridsh Joumal of Cancer (1998) 78 (10) from the population of interest that had been labelled in nitrogen. This has the advantage of assessing the nitroreductive capacity of cells in the location for which the oxygen tension is estimated.
However, the twofold decline in grain density across the rim ( Figure 1B) might well be a consequence of the death of many of the cells during the exposure to anoxia. A similar decline was not seen in fragments of glioma tumour tissue labelled in anoxia (Parliament et al. 1997). nor in EMT6 spheroids (Franko et al. 1982). where additional cell death during incubation was not evident. Furthermore, this decline was not evident in all M006XLo spheroids labelled in nitrogen. Because of this uncertainty, the estimates of oxygen tension based on the nitrogen data are considered to be relatively unreliable. Nonetheless. the good overall agreement between the oxygen tension estimates obtained with the two independent methods of normalization (Table 1) enhances confidence that statistically significant differences between different passages of the same subline (based on non-overlapping 95% confidence intervals) are genuine. Spheroids of clonally derived sublines of M006XLo exhibited a wide range of oxygen tensions at the edge of necrosis. as derived from the labelling data ( Figure 3. Table 1). In four sublines this wide range was obtained from spheroids grown together in the same flask. which indicates that these different characteristics are unlikely to arise from stable genetic variants that co-exist in the parent line. Further support for this conclusion is provided by the xenografted tumours grown from those sublines that gave narrow distributions of oxygen tension at the earliest passage studied. Although the distribution of oxygen tension tended to remain narrow in subsequent passages grown as spheroids. even when the mean oxygen tension varied. the xenografts consistently gave a wide range of grain densities at the edge of necrosis. It is difficult to reconcile this behaviour with an inheritable trait located either in nuclear or mitochondrial genes. It seems more likely that at some point during the transformation or progression of the original tumour that gave rise to the M006 line, an element in the regulation of metabolism was disrupted. allowing the cells to adopt a wide range of phenotypes that are quasi-stable under the culturing conditions used here.
The profiles of grain density in Figure 1 are related to the oxygen tension profiles according to the calibration curve ( Figure  2). Thus. the results from a single growth flask of M006XLo spheroids ( Figure 1) appear to span the entire range of oxygen profiles reported for different types of spheroids. In those spheroids with high inner grain densities. the oxygen concentration profiles appear to be similar in shape to those measured in a wide variety of spheroids using microelectrodes (Mueller-Klieser and Sutherland. 1982: Sutherland et al. 1986: Carlsson and Acker. 1988). Spheroids with a high oxygen tension at the edge of necrosis have also been observed (Carlsson andAcker. 1988: Bourrat-Floeck et al. 1991). similar to those M006XLo spheroids with no detectable rise in grain density. The two patterns of labelling diverge at between 110 and 130 gm from the surface ( Figure IA). As the oxygen diffusion distance into spheroids depends approximately on the square root of the rate of consumption (Franko and Sutherland. 1979). the rates of consumption must differ by a factor of at least 2.5. Some spheroids of subline 13s (p3fl ) showed no rise in grain density in rims that were more than 350 im thick ( Figure 4B). suggyesting that the rate of oxygen consumption in these spheroids was approximately 15% of the rate in the spheroids in Figure IA with high central grain densities. It is likely that the differences in oxygen consumption among these spheroids were expressed by the majority of cells throughout the rim of viable cells in each spheroid: otherwise the rate of consumption would be required to fall to near zero in some spheroids at depths greater than 100 gm. Some of the difference in consumption might be accounted for by differences in the proportion of extracellular space. but a difference in cell packing as small as 50% should be readily detectable (Durand, 1980). and none was evident.
It is well established that cells may consume oxygen at a lower rate in spheroids than in monolayer culture and that the rate of consumption may decrease with distance from the spheroid surface (Freyer et al. 1984: Sutherland et al. 1986: Carlsson and Acker. 1988: Freyer. 1994). In three spheroid types this has been shown to result from down-regulation of mitochondrial function (Kunz-Schughart et al. 1997: Freyer. 1998). The mechanisms by which this might happen are incompletely understood. although substantial progress has been made recently in understanding the regulation of energy metabolism and it is now clear that many potential points of regulation exist (reviewed in Poyton and McEwen. 1996).
Differences in the rate of oxygen consumption are insufficient to explain differences in the central oxygen tension among spheroids. because cell death probably occurs only when the total rate of energy production by oxidative phosphorylation and glycolysis is insufficient (Hlatky. 1988). Changes in glucose concentration in the medium have been shown to affect the rim thickness appreciably in several spheroid types (Franko and Sutherland. 1979: Tannock and Kopelyan. 1986: Acker et al. 1987. In EMT6 spheroids the thickness of the viable rim is affected by both glucose and oxygen levels in the medium in a complex. interactive manner (Freyer and Sutherland. 1986). Extensive studies of these spheroids demonstrated that the severity of hypoxia at the edge of necrosis varies with the size of the spheroids and can be altered substantially by changes in glucose and oxygen concentration in the medium (Mueller-Klieser and Sutherland. 1982). The interactions observed between these nutrients suggested that EMT6 cells in spheroids can adapt their metabolic rates substantially to different supply conditions (Mueller-Klieser et al. 1986). as well as to the concentration of lactate (Bourrat-Floeck et al. 1991). On this basis. it is conceivable that the substantial difference in rim thickness between two sublines with a high oxygen tension at the edge of necrosis ( Figure 4B) might be related to differences in glucose consumption or lactate production. Differences in the glucose diffusivity might also contribute to differences in the diffusion distance of glucose (Casciari et al. 1988).
A novel feature of the present work is that the substantial variations in rim thickness and oxygen tension at the edge of necrosis arose in spheroids of a similar range of sizes grown under essentially identical conditions. We are aware of only one report of similar phenomena. in spheroids derived from sublines of V79 Chinese hamster cells (Durand. 1980). In that case the differences among sublines appeared to be stable with passage in monolayer culture.
The variable presence of hypoxia adjacent to necrosis in xenografts of two human glioma cell lines. but not in a third. was interpreted by postulating that some lines may modulate respiration in response to moderate degrees of oxygen restriction (Parliament et al. 1997). In the present work. we propose that those spheroids that exhibit hypoxia adjacent to necrosis can be regarded as functioning as oxygen regulators. The death of cells in this case is readily understood in terms of the joint oxygen-glucose deprivation model (Hlatky et al. 1988). Recently we found that monolayer BrSish Joumal of Cancer (1998) 78(10), 1261-1268 0 Cancer Research Campaign 1998 cells of the M0059K and M006 cell lines can behave as oxygen conformers (Allalunis-Turner et al, 1998); lending support to the hypothesis that those M006XLo spheroids that have a high oxygen tension adjacent to necrosis can be regarded as functioning as oxygen conformers that reduce their rate of oxygen consumption at relatively high oxygen tensions. In this case we hypothesize that throughout much of the spheroid rim the rate of oxygen consumption is minimal and the cells obtain most of their energy from glycolysis, and that cell death occurs when glucose becomes limiting. Thus, the differences in rim thickness found in putative oxygen-regulating spheroids ( Figure 4B) may reflect differences in the rate of glucose consumption. As both oxygen-regulating and -conforming behaviour clearly occur in glioma models, further study of how this behaviour is co-ordinated appears to be warranted. We anticipate that this model system provides a unique opportunity to study the co-ordination of energy metabolism in tumour cells.